Rise of AI Summit 2020 · Berlin · Applied AI Day | Hybrid Area

AI Solutions: Startup Showcase

i2x, AVA, micropsi industries, neurocat, Solvemate, Merantix Lab, Brighter AI, Cognigy, Jasper Masemann

At Rise of AI Summit 2020 in Berlin, moderator Jasper Masemann (Partner, HV Capital) hosts a two-and-a-half-hour startup demo session in which founders and executives from ten German AI companies pitch their products and take short audience Q&A. Nicole Büttner presents Merantix Labs as an external machine learning department for corporate NLP, computer vision and forecasting projects. Stefan (neurocat) presents automated robustness, safety and explainability testing for deployed AI, including an 'attack generator' for stress-testing models. Alexander Stojanovic presents AVA, a data-standardization platform for security, safety and risk information used by organizations such as London's police and DHL. Micropsi Industries and Wandelbots present software for training industrial robots from camera and sensor data without conventional programming. i2x, Solvemate and Cognigy present conversational AI for customer service, ranging from real-time agent coaching to self-service chatbots and voice automation. A privacy-technology founder presents 'AI for Privacy', a case-based tool for automating GDPR risk assessments, and Clemens (Sia Search) presents a metadata-search layer for large unstructured computer-vision datasets. Each pitch covers the product, target customers and a short Q&A on differentiation and roadmap.

Key takeaways

  • Merantix Labs positions itself as an outsourced machine learning department, citing legal, pharma and manufacturing case studies rather than a single vertical product.
  • neurocat automates AI robustness and explainability testing with an 'attack generator' that runs far more stress tests than manual methods.
  • AVA normalizes security, safety and risk data from many sources into one common, patented standard used by London police and DHL's global security team.
  • Wandelbots and micropsi Industries let non-programmers teach industrial robots by demonstration or camera-driven control instead of conventional coding.
  • i2x, Solvemate and Cognigy each automate customer service differently: live agent coaching, self-service deflection, and no-code voice/text virtual agents.

Detailed notes

Merantix Labs: machine learning as an external service

Nicole Büttner, CEO and co-founder of Merantix Labs, describes the company as an external machine learning department that helps corporate partners with concrete NLP, computer vision and forecasting problems rather than selling a single vertical product. Named examples include automating legal compliance review for a 'magic circle' law firm, classifying customer product reviews, flagging new pharmaceutical side-effect reports, building skin-cancer classifiers from dermatology and histopathology images, and helping manufacturers translate product sketches into production specifications. She positions Merantix Labs against traditional consultancies by pointing to its engineering ('tech DNA') rather than presentation-building background, and describes an 'AI canvas', developed with the University of St. Gallen, structured around business purpose, organization and technology/data pillars to help companies move machine learning from prototypes to sustainable, scaled deployments. She also describes building cross-company 'data ecosystems' in specific industries, such as automotive supply chains, so partners can create shared value from linked data.

neurocat: robustness and explainability testing for deployed AI

Stefan, a co-founder of neurocat, presents a self-funded, bootstrapped company built to address robustness, safety, security and explainability gaps in deployed AI, motivated by real-world failures such as chatbots developing unintended behavior and a UK court case over an unexplainable trading AI. neurocat's testbed can automatically generate very large numbers of adversarial 'stress tests' (described as billions, versus roughly 50 to 200 in earlier manual approaches) through an 'attack generator', and pairs this with an explainability tool offering local and global explanations. The company works across the automotive certification and insurance chain and cites Volkswagen and BMW as customers, plus new projects in telecom and explainable AI consulting, and says its methods are input- and industry-agnostic, applicable to speech, vision or other model types.

AVA: a common data standard for security and risk

Alexander Stojanovic presents AVA, founded after he witnessed a car-ramming attack near a conference in Austin in 2014 and concluded that available sensor and social data could have warned people in advance if it had been connected. AVA ingests incident data (crime, fire, accidents, disease outbreaks, terrorism, social and political risk) from customer, public and personal-ownership sources and normalizes it into one common, patented data structure with a unique ID per event, enabling organizations to share risk information without sharing underlying data. Stated customers include the police of London and London City Hall, public transport operators, and DHL, whose global security operations center uses AVA to monitor risks to its logistics network worldwide. Stojanovic argues the value lies less in individual AI techniques than in establishing AVA's common standard as infrastructure other systems build on.

Robotics: micropsi Industries and Wandelbots

Speaking virtually, micropsi Industries' presenter describes plug-and-play control software that drives industrial robots directly from camera and force-torque sensor input for unstructured tasks such as picking and reorienting wine glasses on a conveyor, work that traditional pre-programmed automation cannot handle. He cites falling cloud training times (from about 16 hours a year earlier to roughly two and a half hours, with a projected half hour) and says a customer can go live within roughly a week once physical setup is in place. Christian, CEO of Wandelbots, presents software that lets non-programmers teach industrial robots by demonstration with a tracked handheld device, generating code for multiple robot brands (KUKA, ABB, Universal Robots). He cites a Volkswagen end-of-line inspection use case needing only 20 to 50 reference images per check point to reach about 99 percent accuracy, and a demo in which Microsoft CEO Satya Nadella remotely taught a robot in Germany a sealing process from Redmond. He says less than one percent of manufacturing jobs that could be automated currently are.

Conversational AI for customer service: i2x, Solvemate, Cognigy

Michael, founder of i2x, presents real-time speech-to-text analysis and live coaching for call center agents, delivered to the agent's screen in under half a second per cycle, alongside customer-satisfaction and emotional-tone analytics; he cites 10 to 30 percent conversion increases and about 50 percent time savings for team leads, and frames the product as agent augmentation rather than replacement. Eric, founder of Solvemate, presents a self-service chatbot platform built on what he calls 'dynamic decision trees' rather than pure NLP, aimed at deflecting routine requests for companies with 10 to 200 service agents; he says setup can go live the same day and cites Gartner's AI hype cycle to argue chatbots are now past the hype phase. Silas, from Cognigy, presents a low-code/no-code platform for building voice and text virtual agents across more than 30 channels, offered as SaaS or on-premise (on-premise being the preference of GDPR-sensitive customers such as insurers and German companies), citing Lufthansa as a client and reporting around 30 percent call deflection.

AI for Privacy and Sia Search: governance and data tooling

A privacy-technology founder, a lawyer by training, describes building the 'Privacy Suite' after his team spent 70 hours on a single GDPR data protection impact assessment; the tool now automates risk scoring and, through 'AI for Privacy', uses case-based reasoning across anonymized processing activities to suggest risk mitigations, funded in part by an innovation grant from the German state of Hesse. Clemens, from Sia Search (CSH), presents a metadata layer for large unstructured computer-vision datasets, arguing that engineers can spend up to three-quarters of their time on manual data handling; the product uses interval-based metadata matching, which he says is roughly three orders of magnitude faster to query than conventional per-frame databases, and is offered free for academic use on the nuScenes dataset.

Quotes

We're basically your external machine learning department for companies.

Nicole Büttner

There is no minute passing without somewhere in the world stuff happening.

Alexander Stojanovic

The best customer service is no customer service.

Eric (Solvemate)

Questions answered in this talk

What does Merantix Labs offer companies building machine learning products?

According to Nicole Büttner at Rise of AI Summit 2020, Merantix Labs works as an external machine learning department, building bespoke NLP, computer vision and forecasting solutions for partners in fields such as legal compliance, pharmaceutical safety monitoring and manufacturing, and helps organizations structure their path from prototypes to scaled, sustainable deployment.

How does neurocat test the robustness of AI systems before deployment?

According to neurocat's co-founder at Rise of AI Summit 2020, the company's 'attack generator' automatically produces very large numbers of adversarial stress tests against a model, far more than manual testing could achieve, and pairs this with explainability tooling to give local and global explanations of model decisions, used in the automotive certification and insurance chain.

How do Wandelbots and micropsi Industries let non-programmers control industrial robots?

According to the presenters from Wandelbots and micropsi Industries at Rise of AI Summit 2020, their software lets users teach a robot by demonstration with a handheld tracked device, or drives the robot directly from live camera and sensor input, generating robot-specific control code without conventional text-based programming.

Mentioned

Robustness testing, Explainable AI, Attack generator, Conversational AI, GDPR data protection impact assessment, Dynamic decision trees, Metadata search for unstructured data, Plug-and-play robot control, Case-based reasoning

Keep exploring

More from i2x

More from Rise of AI Summit 2020

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 23,725 words

Show the full transcript

hello and welcome back to the ai hybrid stage of rise of ai we're now talking about ai solutions and we have a nice lineup for you of 10 companies that are using ai in their everyday work improving their customer products and i will probably have have you here for about two hours so i hope you have eaten well you're prepared you've been to the toilet um and starting off with something very very uh important here in the berlin ecosystem i think we had yesterday adrian locher we had a rasmus order today from both from marantix and i'm very glad to have nicole bittner here dr nicole bittner sorry nicole who is the ceo and co-founder of mirantix labs and she does everything related to ai for those other two so she actually knows how to do it and she will tell us now what the runtik's labs is doing how they are applying ai you call you have 10 minutes and then we do q a fantastic thank you so much i'm actually not dr

nicole but um we always call rasmus dr ai so maybe that's that's where that's coming from thank you very much for your kind introduction i want to talk to you about mirantix labs we're basically your external external machine learning department for companies and we often um yeah oh little glitch exactly so um we are basically the portfolio company of Merantix that brings the the mirantix knowledge on data-driven business models and machine learning into companies right magazines and adrian have talked a lot about us we are basically um not building a product company that we are um that is focused on one industry vertical but we are a service company that works together with our partners from different areas on their concrete machine learning issues and and challenges and also opportunities i think yeah you've already seen what an amazing team we have i think it's important to say that the expertise that we have spans mainly three domains we are very heavily specialized in natural language processing and also in computer vision and are rapidly

looking also at the forecasting space and reinforcement learning and i wanted to bring to you maybe a few examples of what we do there actually so in the natural language processing field right we for example work with one of the magic circle law firms and help them automate legal compliance topics right these are often issues where uh it's it can be tricky maybe to um to actually employ a product that you can find in the market and it's a little bit easier for you to work with a service provider that builds something that's bespoke for your systems and that really can be integrated very well in existing workflows as you know the legal domain is quite specific and also quite special in its in its workflows we for example also look at fields like customer product reviews um this we do for example for consumer electronics platforms and we also for example work in the pharmaceutical sector where they are trying to sort of automate how to deal with new side effects and new alerts

for side effects in the computer vision department i think my colleague claimants will also talk about cs search later on today and about the great solution that they have in the computer vision department we for example focus on things in the computer in computer vision some medical so medical use cases so we for example look at derma histopathology samples and basically build classifiers on how to easier diagnose skin cancer we also are for example working on products and like end-to-end solutions to look at the future of for example in warehouses how to look at returned goods um when you're when you're thinking about that whole supply chain and i'm very happy that we're also very active in the forecasting space so we're for example working with manufacturing companies this is very widely applicable to a lot of smes and producing companies all over germany and europe where you have to basically take a product sketch or prototype sketch and actually translate it into real production environments and a lot of the companies we're working with

are facing shortages and qualified labor in this space and we're very happy to assist them with partly automating this approach and helping them basically produce these parts much quicker and translate these these sort of mach these part specifications much quicker to actual production descriptions so how does a laser or how is a specific part sort of produced from a metal block for example in subtractive um case um yeah so i mean these are some of the cases that i talked about i think one other case that's really interesting as well is optimize design of experiments so there we for example work in the chemical industry and help looking at how to make the r d efforts more efficient so basically reduce the number of experiments you physically have to run to find a new composition for a new material um which is which is very interesting and also um applicable for example in the former pharma sector so those are for example things that we do i think you will hear about a lot of

cool solutions today and right a lot of companies out there i know some of you are making ai makers some of you are talent some of you are actually corporates and we are here basically renting lab's mission is basically when you when you think about should we buy or make right should be outsourced something or should we build it in-house that we help you bridge bridge that gap i mean and also make some of the companies more ai ready we're working with the university of san galin on this together on a basically ai canvas as we as we're seeing at the moment a lot of companies are experimenting around machine learning use cases so they're already trying out things running prototypes building pocs but it's really challenging for some of the companies to actually scale this and translate this into sustainable business models into a really long-term uh deployment and high sort of uh scalable business value and that's why we've we've decided to build this canvas and to help companies guide basically navigate this

process and really harness the full power of ai in a very systematic fashion focused on three pillars one is the business pillar that you see in green um so set out with it with a clear purpose the second one is focus on organization how do you build and cultivate teams and setups in which you can actually also quickly build and iterate on uh machine learning products and the third pillar is technology and infrastructure and data so what is actually a good setup in this fast-paced world of moving tools to actually set yourself up for success and these are this is sort of one tool that we where we really have a high ambition to help existing organizations to really um understand uh and and really help them to harness the the full value of machine learning models um the second the second passion i would say right is also around this didn't load that's not good so i don't know i will describe it i will just stay on the canvas line so a second

thing that we're doing and i know rasmus has talked about our campus as well is we believe that to really transform um business models right you have to think also in your strategic supply chain you have to think on how are you actually as a company integrating in this whole universe and your whole industry and maybe even into other industries and how can you actually sort of lift your lift your gaze away just from the from the data you're producing to kind of what is the actual value creation process of of your whole industry of your whole value chain and to this purpose we also create data ecosystems and ai ecosystems also on our campus in berlin together with companies who belong to the same ecosystem so one could for example be car manufacturing right you could imagine a lot of different companies in the value chain of car production from tire makers from from infrastructure providers right all around this sort of topic of mobility working together and creating impactful use cases so it

actually we're starting to to to think about this talking to tire manufacturer who have some very valuable data on what they're doing in their tire production but how much more impactful would it be to actually link this data to other data points that are coming out of car and really create new value propositions and use cases to the driver to infrastructure operators etc and for this we have to i think very strongly thinking ecosystems of data and also value creation same could be for example applicable for a domain like building or or construction right so yeah one message we have a very strong machine learning and data-driven business model competency we want to empower companies second message to harness this and to really make themselves ai ready and become data-driven companies themselves and the third thing is we're here to build a data and data-driven machine learning ecosystem and bring together different players from industry and from the startup world i think i'll make a full stop here and wait for some of your questions

thanks a lot thanks a lot nicole and sorry uh for already promoting you i'm always i'm still dreaming of my doctor so that's maybe the case um so thanks a lot um one question uh maybe coming like a typical venture capital question sorry but since the space of let's say let's call it ai consultancy and please shoot me for that word but it's getting more crowded and crowded bcg has won mckinsey and many others what's the special position of of mirantix labs or special edge maybe as you might call it i think and thank you for this question actually so i think i mean i would say uh traditional consulting companies have a lot of strengths right their traditional strengths that they're coming from fundamentally though i would say they're not organizations that are built to deliver code i think they're moving there right they were built to verb i don't mean disrespectfully powerpoint presentations very impactful ones and inside industries we're truly a tech we truly have tech dna and we know how to

build tech companies right and when you look at um when you look also at retaining talent and staying on the edge of these technological developments i think marantix has a much better starting position than some of these bigger consulting companies they have probably other advantages and i wouldn't discard working together with them but i think we have different strengths that we play too and i think that you touched a very important point what i see in in our portfolios the larger the companies get with their ai models and products it's getting quite boring uh for at least data scientists because it's about stability it's not about new things so who is who who are you looking for at mirantix labs what kind of people yeah so i mean we're looking um we're definitely looking for people who want to stay at the edge of research right we want to be pushing that frontier and we're looking for people who basically want to see different problems and think about the solutions and conceptualize those solutions and

we then help companies right implement that and deploy that and i should translate that into a valuable product or solution for them ultimately however i think you know also that companies want to and also have to build midterm their own capabilities right in-house so you know we will always be kind of pushing at the frontier and now you can say it's a push or pull thing right depending on whether we're in front or behind or maybe sometimes from both sides um but trying to keep basically companies on their toes bringing them the newest um research insights and translating that into actually workable solutions that you can actually use when you're you know working in your in a back office job or when you're a truck driver when you're you know working in a law firm so final question continuing on that point um so what is on the edge so let's look at the next year what do you see will happen in ai maybe in terms of new models new approaches what is the

edge i mean i'm sure a lot of people have touched on this open ai framework we see a lot happening in the nlp space and i think you know open ai new um gtp3 is really impressive comma however and you know how are you actually going to use this as a company so with this with this fascinating technology and right and also this this breakthrough you actually have to find real applications i think two and and i would say two three hurdles that are there that we that we're looking to solve technologically are for example um small data sets so how do you actually apply these great models right i mean gdb3 was traded under 175 billion parameters so that's kind of i would say qualifies as large how do you actually make this workable when you're in a company and you know you don't have this kind of amount of data right and that's actually the challenge that people have you know you see this in theory oh yes supposed to look like this

balance data set large and so on and then you look at your own data and you're like so for example when we look at working with law firms right that's for example one of the challenges you have you have very small data sets or maybe low incidence problems so looking at synthetic data for example i think will be highly interesting and how to how to create valuable synthetic data sets around specific domains and not just sort of scrolling the internet that's one thing and i also think in manufacturing actually this example that i looked at that i was presenting sort of in this whole design of experiment reinforcement learning space about how to reduce the number of attempts it takes you to calibrate a machine for example to do farmer research right these are all really very impactful use cases and i think they will bring a lot of value to a lot of industries cool thanks a lot nicole and everybody who's interested in that please give her a call you can apply your

rise of ai discount code hopefully i guess you guys are not not cheap thanks for being here um and next up on the stage since we spoke so much about ai regulation and also people trusting or maybe mistrusting ai i'm very glad that stefan hintz is here coming from atlas wolf in berlin i just read you recently got accepted to nvidia who are supporting you and sheffield will tell us how he enables he and his team enabled trust in ai you yeah well so first of all thank you fabian and thank you nicole for this fantastic pitch thank you um for this opportunity yeah my name is stefan insert who is stefan insert chevron is one of the four founders from the neurocat i have a strong economical and management and sales background the other two founders have a strong mathematical background and yeah and the fourth one uh was responsible to to pay the bills in the beginning and here and yeah i'm very happy that we are seeing the new cat is growing

here i say we came from the streets that mains that means we we grow up organically so we have no funding um we our first mission was to understand the demand i think um that's very very important experience to understand the market and yeah and our mission is to enable quality safety security for ai very proud to have tarik now on board as a new cto and a new board member and joachim and matthias a lot of exciting people i think we are close to 40 um i see 36 so yeah and we're working in atlasov atlas hope that means science at work what means signs that work they're doing a lot of science and transform it into an economic in a economy and that means um we we try to bring new stuff into the market and um here and you're seeing we have we already published papers uh cvpr and i think there's now a new ips paper outside we have five uh patents pending still here and um yeah what's our goal

usually our goal is to to solve the key problem and an ai um deployment that is the robustness this trustworthiness and uh safety security compliance and explainable and your that's the nature of itself every startup like to solve problems yeah that is that is how it is and yeah what's what's the problem we felix has covered it three four years ago and we need to think a little bit if you understand ai at this input input processing output and an ai is only so good as its input and what happens if the input is being manipulated or sensors are not working uh properly or noise is present in the environment i tell you what happens that and we already know that and that is not a funny story we know people lost uh the life and yeah um and this most safety security problem a robustness problem and you have and i have other stories as well so maybe you remember um in the past um there was a chatbot from facebook he developed his

own language or the chatbot from microsoft became racism i remember last year in london there was the first uh court file it was a trading ai and this trading ai the users don't understand how this ai will work and they lost money and your and the reason was if you don't understand the system how you can cover the risks from your clients and so simple is that all these challenges we tried to solve yesterday we have no reliable test beds we have close research solutions to analyze the ais we have no really methods how we can solve the robustness problem how we can understand the decision boundary and um a lot of black fields in this research ai field yeah and now i have the good news of course today we have great solutions and eurocad challenged this research topics and we have now reliable testbed we have the solutions ready to make the ai um robust make them safe to secure you see in the automotive driving we're working strong together with certification industry

with audit industry with operation industry with the developer the whole food chain you know and the insurance company in the end of the day want in risk assessment all these challenges you can solve for solutions industry agnostic it's unlimited robustness stresses it's uh tremendous we solved a robustness problem with um um agnostic to the differences um in the same time we increased the performance all the ai researchers know what that means so we have our neurocad labs we start there we make the services we have our solution to analyze um the models the data is full automatically and then improve it just press the button wait buy a little bit compute and become certificated we have success stories of course with volkswagen together we we enable safety security not only in automotive sector we have much more ml who will be deployed in the course i'm very happy to start next year with an um explainable ai project thank you claudia for your support on this site telecom here we're talking to bmw munich re

they're we working on risky evaluation of course the uh consulting companies are interested in trusting ai here and very strong connected with all the tools to build on certification and one that is and maybe i want to go back you know you need to understand that it's not only a research field that is a high scalable commercial ready to commercial use solution that means you can run million of stress tests and we have ready million of stresses one of the big uh difference between google and eurocad it's stable bam how ma how much money it depends how much money you have so one of the first highlight is our attack generator that give us the uh opportunity to to generate million of stress tests that means in the past maybe you remember we have 50 200 stress tests and with our solution we have bunch of billion of stress tests and so that means we can not only analyze the ai system and the best way we can improve the system in the same way

so that means that is agnostic that is real time you can deploy it on-prem and then you have a safe secure solution no hackers have no chance no noise anymore and here and that's another highlight is our what if um from the explainable ai pillar uh our solution is more flexible maybe you have heard from the google solution we have still a patent on that too you can use that uh um i can say in a brighter way give you local and global explanation here and i'm very happy to see what will come next year so this solution give you an in one point or in in one platform all the tools what you need to become trustworthy to understand your algorithm to improve that and to monitor them if you see um the uh the the slides you will see uh how we are situated in the in the in the food chain of the ai you're doing in the beginning something with your data then you train you know this iteration and here

we have our testbed and in the end of the day during the run time we have our improvement solution i called firewall or explainer 4ai yeah why um came the clients to us why they choosing first of the motivation of the automotive industry they don't want to drive millions of miles they don't want to run the systems for thousands of hours and here we analyze automatically the edge cases enable functional safety be head of the ai compliance curve and yeah um a lot of companies we are working with and yeah what's our roadmap our roadmap uh for the next three months is to to implement uh robustness uh evaluation hey what's so funny you guys laughing there is it's all good yeah okay and we can move to turn to red maybe i'm doing something wrong here so yeah we have a threat model selection here the version two robustness course and finally we enable rnn her architectures to improve uh the robustness and the comprehensively one of our success stories with volkswagen your other

our clients very happy to work with us saying we leaving the fields here you know everybody is saying he is the best so of course we are always the best or already your best how you get that solution it's very easy of course go on the website just register your interest fill up your email and press the button and then you can talk about the pricing later thank you and um we have 40 seconds too early so if you have any question feel free please thanks a lot stefan thanks a lot um so i'm just checking here okay um so first question for you um we spoke a lot about explainable ai we spoke a lot about regulation and understanding those models and you're actually telling us now that you can do something of that uh sounds very revolutionary is it is it two very specific cases that you are solving or what what are you covering exactly so thank you for this great question yeah that's what i have forgotten we have we are

definitely industry agnostic we are input agnostic it doesn't matter is it a speech recognition it's a division system an end of the day you have always the same methods and it's always a matter of the plots how you plot the problem how you will plot the reports but to analyze his methods and mathematics behind is always the same and here and so we are agnostic to everything to task agnostic input industry vertical doesn't matter to us we have a lot of experience since four years and you have a very reputable customers there and obviously they probably also have their own data scientists so of the audience listening who would would be the best customer for you what are the people are the ones that are building their own models are the ones that take more the off-the-shelf ones eric customer customer has a good question first of all customers are king every customer is welcome you know it's very interesting to work with different stakeholders you know in different knowledge level you have sometimes guys

like me who maybe understand the wording ai but then you have technical nerds who can only talk with our technical guys so but usually um the good good client client is a client who was willing to pay and and never say no you always yeah and hopefully it's helpful yeah good um and are you also giving advice how to improve models so are you just detecting something or he's improving yeah that's that's not a good question we transform actually from a service company in a sas company so we in the past of course we was doing a lot of consulting we talked to the research guys and say hey doing it this way and move on this and and um finally now we we have your strong partners like bearing point t-systems uh uh a lot of converting companies who now now help us on this and we just deliver the the test bed and the monitor bed and the improvement solution but that's our mission to get it safe to get it trustworthy and

the right political meaning thanks a lot thanks for being here and i guess if any customer wants to contact you they can use yeah feel free i'm very happy and thank you and again thank you fabian and to the team thank you all right um that was trust in ai next i think that's a very uh nice follow-on topic about this um we have alexander stojanovic i hope i pronounced it correctly thank you of ava and they simply claim or somebody referred to them as the palantir of europe so i think that's about it about enough of introduction please welcome to the stage alexander where is there is a clicker the presenter thank you thanks so much all right thank you thank you for being here so um i'm here to today to give you like a like a quick overview on ava and what we do um could you please restart the timer yes perfect so they have a little bit like the time inside now um before i'm going a little bit deeper

like into into explaining like what what ava is about i would like to quickly show you a video which will give you a little bit of background how all of this has actually started this morning we begin with breaking accident in austin texas as the town hosts the hugely popular south by southwest tekken media conference two people are dead and two dozen are injured after a car plowed into a crowd police say the driver was a drunken driving suspect being chased chased by police so the video is probably not working perfectly but what happened there was actually during my stay in austin 2014 i witnessed actually the senseless death of uh of many people and many more people have been injured uh during this incident um and it was actually on my way back from austin to germany where i figured out that this was not just frustrating but actually mind buggling that uh in a w in a in a country and in a town where everything was connected like austin where the police

cars have telemetry where cameras every everywhere during south by southwest you have everybody like running around like with a smartphone and doing doing videos and pictures still there was no technology there was no bridge to connect all of this information to detect this pattern about what was about going to happen even though the information was there so also after i returned looked at it in retrospective it was possible alone from the data that was actually there you could have prevented like this event this incident this accident actually giving warning to people the only thing that was really missing was this bridge so um being there in ex and and and and coming back like i sat together with my co-founder sasha knop and we we set off to deploy cutting-edge data technologies and ai to solve this particular problem and to create like a data fusion platform and artificial intelligence for a safe world by tackling this problem we actually very quickly came to the conclusion that this is much more than just about like

creating a set of technologies that connects ai methods for example natural language processing and pattern recognition and machine learning but this was really about epistemology or as german would say like a chemist here we really had to understand what is happening when we're talking about security safety and risk so what we do effectively is to look at incidents events things that are happening in the real world we are parsing and taking in this information through various channels and the core aspect of what we do is to put all of this information into one common structure this means that regardless of whether an event an incident like an ant accident or crime or terror attack or a natural disaster comes through one or the other media or to the one or the other channel our system needs to understand all of the all of the underlying knowledge in it so when did it happen what happened where did it happen how reliable and trustworthy is the information that we are seeing there and that we have

to put into one common standard regardless of the source that we are looking at and while we have solved that during an r d project of almost five years before we actually started to work together with customers in markets we created this this uh this this awesome opportunity to actually bring this information all around security safety and risk into one common standard that defines all of the things that are related like to this type of information this is unique we have patented it now it is actually patented it's not not even pending anymore and based on that we can create solutions and these solutions and interfaces are what our customers currently are using and doing so before i go to that just quickly a quick introduction of what type of sources we we intake so first of course these are information from our customers customers that actually have incidents accidents things that are happening in their organizations in their cities in the in the places that they feel responsible for we have public ownership data

like open data from cities but also closed up information from police forces and other authorities we have personal ownership this can reach from social media and sensors over of course like to professional intelligence providers broadcasting media social media all of these kind of sources they all move into ava they're all translated into this one common standard and made available we do this for aspects such as crime we do this for aspects such as fire come on accidents catastrophes social and political risk which is an important aspect and of course health risk so um we've been very very early to see what was going on in the world when the pandemic or what back then was still declared like an epidemic started like in wuhan and uh we are supporting like much many of our customers like in this aspect and of course terrorism is one aspect today this is not just a presentation or pitch this is actually existing ava is up and running technology that is scalable and very very reliable and resilient robust

to use by some of the most advanced organizations in the world amongst them we have the london city hall and the police of london which uses our systems to actually improve the security safety of their citizens we're doing some special projects with them aside for example in regards like to organize crimes things that you would in germany refer to as clan criminality for example um for public transport operators we do similar things so you can imagine that a public transport operator of any country is confronted with a lot of things that are happening in there in that public transport network not only at train stations or within uh within trains but also along the tracks and uh when you have like for example severe weather or those kind of things so helping these organizations to actually understand what is going on to figure out like from the history what has happened and to figure out what could that mean for the future to get to a point like of a of a predictive scoring in

that regard and we do that for large corporations or ngos or any other uh type of other large organizations one of the customers that we are working very closely with is dhl worldwide so some people don't really quickly understand like how important security is actually for the operation and the resilience of this organization but you can really uh um imagine that dh if dhl stops moving for whatever reason because of some incident or some other severe security or risk-related event the world stops moving right like there's no way that this world operates that we the way that we know when organization like dhl is not operating so we are helping the global security operations center and the team there to actually oversee everything that is happening on the world ranging from piracy in the gulf of ghinea over to border closing somewhere in southeast asia so everything that actually could pose a risk to them and they are saving a tremendous amount of time and resources by actually doing those kind of things fully automated

with the help of ava and of course like trained by their own analysts and working like with these teams very closely together you will find us in security situation centers so this is for example how this should how this looks like this is actually an actual view like into our operator's interface like for professionals we are available in the field so also field personal can use like our technologies and ours our interfaces to access this information on the go so for example myself i'm always like best prepared so when i run like through berlin and something happens in my vicinity my smartphone uh rings up and tells me hey listen there is a protest going on short but uh close by or there's a large fire or any type of incident so that of course is something that has also been used like by professionals who also use like our interfaces to to uh to track the information that they that they see themselves so if you imagine like a large public transport operator with

eight thousand nine thousand security staff um they are using ava not only to get information on how to best like prepare themselves and where to go when they're having like their shifts but also to to capture information that they're seeing while they're doing and all of this has been trained uh further like in in our system and the fact that we're doing this really on this basic level really at the core defining really a data standard and a core standard on how to manage and distribute and collaborate over security related information gives us the opportunity to do things that were unimaginable before so some examples for that of course is your car which in the future will tell you don't park me here this night because there's a higher risk for vandalism or take a different route because there is a there is a risk like for you for theft over carjacking which of course is not a problem here in berlin rather like in other places of the world but you have so many

other use cases and business cases which heavily rely on security related information just to name a few insurances that underwrite and do like security and risk analysis you have quant funds we have been approached by some of the largest financial institutions of the world which are interested like in alternative data they want to know 30 seconds before cnn when there's a coup d'etat going on in turkey so they can they can hatch with leveraged uh uh with leveraged uh um activities against the turkish lira under the turkish stock market like things that we haven't even imagined which could have been possible are now possible like through the technology and the standard that we have created and that our customers are using to actually uh build their ideas upon now eventually we will get to the point where we get back like to where it all started like to create something that is available really to everybody everywhere in the world like really a personal guardian angel something that tells you when your situation for yourself

for your family for the police for the people and the places that you love is okay and how you should behave so with that i hope i gave you like a very broad overview and an interesting insight into ava thank you so much thanks a lot alexander thank you thank you um yeah i ran into that protest by the way i didn't have your app so i was almost late for the conference um all the use case you have shown they're very broad they're very impressive very broad so maybe because i read the ipo filing of palantir sorry i have to refer it again and please correct me if you're slightly different but there is a core technology underlying to all of this that's why i imagine you have these broad cases can you explain that a bit more since we're an ai conference what is the core of it so the core of it is actually the part where we take information and around incidents things that are happening security safety and risk related

and put it like into this one common structure and standardize it not only in terms like of its categorization and the indications everything that relates around that but also we give it we give it like a unique id so you can actually talk about the same thing regardless of which organization you are and this makes total sense if you think about it like in berlin you have the case that for example at the train station in berlin you have a you have a consumer electronics store so the security people of this country electronics store have no clue that there is a train running up like with hooligans in there right so they're not prepared whereby on the other side uh in hamburg altona at the consumer electronics store of the same chain uh the store manager decides to do an autograph hour with local rap giants so they all of a sudden i have in the middle of the rush hour 500 600 teenagers that are rioting that want to get in and the station

manager doesn't know anything so by creating this common standard this one common standard in the middle we're actually creating for the first time the opportunity for organizations to collaborate with each other without actually giving away like their own data without giving away their own information and by putting that in the middle this is really like the like the like the standard or the platform if you want if you don't want to look at it from a technological aspect creating all of that of course is very technological and very ai driven yes and it seems very relevant because what i learned is that you it's mostly easier to have very specifically trained models for every case or every location so you have a general model that you train for the local one okay and do you uh do you use some modern forms of transfer learning is there anything that makes gives you an edge there because everybody else could follow you take the same data do the same thing i imagine there were there was

there was of course in the beginning like the conversation uh when we talked like really about the data product right like offering ava not only like through interfaces but also as an api right so if you have access to the api couldn't you just copy it and create the same product i think aside like of all of the ip that we that we have secured i think the biggest issue around that is that there is an ever ever ongoing stream of stuff that is going on there is no minute passing without somewhere in the world stuff happening and i mean we see it right like ava is detecting the rocket smashing into a building in idlib the stabbing that is happening right now at columbus circle in new york the severe weather event that is happening somewhere in philippines of course you can build that as well but i would assume that once we have established ava really as the standard also like as the communication and exchange standard that we are we are facing

we are really facing like like a like as the west coast tech uh community would say like a winner takes a dollar situation there is no multi-homing it's not like i get netflix and amazon prime you will get ava or anything else like ava but you won't run like two or three or four systems aside so for us deploying these technologies being fast with that is important but even more important is to actually uh is to actually establish ava as the as a standard and as the as the structural uh nucleus around like where all of this information worldwide is uh is layered around sounds like a very very nice mission thank you for that alexander thank you so much thank you for making this world a better place thank you thank you so much that's exactly why we started thank you guys for cool thanks hi all right we're continuing now um with a continuing now with a virtual um virtual presentation from ronnie rooney who is um the ceo of micropsy industries and

we're now taking talking about a safer world ronnie can also help with that and i think he has invented the most sensible robot or at least the software that makes every robot a robot more sensible so i'm very very glad to welcome ronnie and please go ahead hey thank you uh so yes i'm going to talk about um plug-and-play ai for industrial robots um and um wha why are we doing this uh and why it's inter why is it interesting right industrial robotics is something that happens in factories and we normally do not tend to think about you know what goes on in factories too much the the larger perspective is that factories production demand automation they have done it for commercial reasons for a very long time it just pays to automate production in times like these you also see that it just makes production more resilient with just you know in this year for the first time i think in my lifetime experienced scarcity of certain goods um and when you have scarcity

of certain goods it's either just the delivery that's the problem or actually production and that was the case this time and it was partly just more demand was also partly it was hard to run factories because people couldn't come to work or didn't want to because they felt unsafe so you know apart from just making production more resilient and fulfilling a demand for being more effective through automation the really large perspective is that automation in the west where we have aging shrinking populations is one of the very few remaining sources of growth and that means higher living standards so when we talk about um universal basic incomes uh the work still needs to be done and the reason we can even talk about it is um that more and more work goes to technology goes to machines goes to automation so it's really a secular mega trend and it is also what makes our life so nice and are you know makes producing in economies like the german one or the us one where wages

are high even possible to this day and the problem with automation is it's also extremely hard um and it's it's extremely expensive and you you know put a lot of hardware in place which also makes it uh very inflexible so we're in a situation that is comparable to um i.t in the 1980s right so there are these big industrial players and they have computers and it's quite clear everybody should have computers going forward um but you need to buy them from ibm and people in white coats come and you know it's it's a project and then you need to have experts running it and it really feels like this is unsolvable this is an unsolvable problem the technology is just too complicated the technology is too far out and too sophisticated to be run by small companies who are supposed to do something else right if you're an office some people in this room might recognize what this office is there so if you're if you're an office you don't want to deal with computers

you want to deal with you know catching criminals in this case so um the change came for it when it went from you need to buy it from ibm and understand what you're doing to you could just buy a pc a small product that encapsulated what you needed to have about the technology that was easy to deploy and it was easy to run for you and wasn't at the standard of the big guys but it was good enough so you could do it yourself and that's where we are with automation particularly robotic automation these days you think about what a factory is um so that you like the basic model of a factory is you put stuff in on one end um and the factory reorders the stuff and the product uh comes out on the other end and the product that comes out is more valuable than the stuff you put in right that's what that's what factories do so they're really about flows of material raw materials or pre-products that you know goes

for a number of steps and value is added in every step ideally and it is very hard to make every one of those steps as the material flows through the factory predictable right so if you want to handle it you want to manipulate it you want to drill in it or whatever you're doing you need to know where the thing is and that is expensive because you need to control every single atom of the material as it flows through your factory it's too expensive it's too inflexible people don't do it so what they do instead is they use humans because that's much easier easier to set up and it's also cheaper and this is like the lever on which you can operate with ai because you can bring the flexibility that you need in factories through ai to actually start replacing human make them more resilient make them more effective so if you think about this in the context of robots what you want to do is you want to find material that has somehow

been delivered to a workstation and you want to transfer it or put it into a machine or somehow manipulate it and what you want to do technology technologically is just control the movie you're making with the robot directly from the camera image that you see now that has that has been hard historically and the only reason why it hasn't happened so far is that it is so hard and the only reason not to implement it is that you can't and we've solved that over the course of the last year what you see playing in your stream right now is a robot picking up wine glasses this is a real use case somewhere in northrend is fine that where this needs to be done and they're sort of semi-ordered um on a table or in the real factory on a conveyor belt um they have been you know just put in some place and the robot is supposed to find them um turn them around and put them into some order so a cardboard box can

be put on top of it and this is what we're solving here so the robot really makes a search movement and finds these glasses and this has not been solved by you know some experts in berlin this is microbes industries this is a product um that knows how to learn these things and has been trained by somebody who does nothing who doesn't know anything about ai they know a lot about class to find these classes and perform the task i'll let it catch the last one before i move on yes well done so um you couldn't come up with a use case like that on a whiteboard in berlin right finding turning around wine glasses isn't a large use case that robotics or even automation used to deal with robotics you know predefined automation typically happens in automotive revolves in the picture this is where you do welding and painting and palletizing you have a lot of installations it's a very well understood problem and it's been sold for 40 years um and the real

value the interesting applications are things like wine glass picking things that nobody know knows about before you actually go to the factory and talk to the people and understand what they're doing there they're completely unmapped they're in testing they're in assembly they're in infra logistics just moving stuff around and these these people all use humans for performing these tasks today because that's their only way of dealing with the disorder that's natural and that doesn't force them to predict where every atom is at every second the automotive oems can do that everybody else needs humans or ai so um as you can imagine if you if you are you know running a factory um what you're essentially doing is you are running a machine that prints money um you feed it uh it does something more value comes out on the other side you sell it you print money if you're going to change anything about it for instance putting a robot in there instead of a human you better make damn sure that it prints

more money afterwards than it printed before you made the change so anything industrial automation is the long-term thing um it's trust building people will use um suppliers that they have been working with for a very long time that they know exactly uh will work and will help them and that they understand you know will deliver the results and that have have technology that is actually reliable and what we are doing is um we are asking to make quite a big change right this is unheard of doing real-time control from sensors in your production we do however believe that this is going to be the default and we're in the process of proving to that market and you can see you know some of the players that we have proven this to and who now believe it from the automation space or from our customer space that this is possible and this is what they can do in the future when they realize okay a human is getting too expensive or it's too shaky or too

dangerous here and we really need to use a robot that's flexible enough to solve that problem if you want to know more about the winegas winegas picking use case here's the url there's some documentation and the video to re-watch thanks a lot ronnie and maybe to start with a with a very simple question the robot we have seen that's a robot that wasn't built by you that's a normal robot yes we don't make robots um we are a software company um you can add our additional control component that's like a i mean among us it's just a gaming pc right it's a piece of intelligent that has been packaged into into a controller you hook up this controller to a robot that you already own or that you you know buy to put it in place in your factory for use case and then we start to control this robot and the movement that this robot makes from sensors that we connect typically a force torque sensor and a camera or a couple cameras nice

and when when you build this in-house in the factory how how long does it take i mean do i have to train it for months not anymore so um just one kpi from uh from you know what we've reached over this the course the course of this year is uh we went from training times so this is cloud training time so you recorded the data you send off the data to the cloud and it gets processed into a capability for the robot we went from i believe 16 hours a year ago to two and a half and we believe we will be at half an hour next year so you know to a degree that doesn't even matter the actual robot interaction time that you need for solving something like the wine glass picking use case went from um it was like four or five hours with a bit of experimenting in in between to less than half an hour again it doesn't really matter the amount of time you spend on building the physical

setup by far exceeds the amount of time you need to you know invest to actually create the ai to make the to to make the movement so to answer your question uh if you've got your hardware in place uh and everything's uh set up and your you know physical factory has already been changed you can be live in a week or even less than that typically it takes a couple of weeks to you know re-plan uh where you put the robot where you put the safety equipment um what group of fingers you will be using um and all these types of things and talking about artificial intelligence that you're using what's actually what what was maybe or will be the toughest part in ai where you created your edge because obviously not everybody is doing what you are doing yeah and that is that sort of that's sort of the you know if you look at just neural network models that deal with visual input what i typically do is classify um and classification doesn't

help right we it doesn't help us at all um when we get feedback that this is a wine glass we need to know how to move relative to it and that completely did not exist we had to create that from scratch there's a bit of research that you could read five years ago when we started and we certainly did that but then at some point we just have to sit down and essentially develop the full stack for moving the robot in real time but also doing this from a neural network that you just do not have a model on github for that you can just use cool so everybody who is using robots at the moment are thinking about them they probably should at least talk to you and figure out if microcan help thanks a lot ronnie for your time and coming up next we have michael brim from i2x and michael once told me that he wants to use his ai to help people on dating i think he evolved since then so

michael welcome can you hear me michael hello hello michael can you hear us yes then shoot in english please fantastic uh thank you very much uh i'm very happy to be here i'm michael founder of i2x um and i'm happy to tell you a little bit about our solution how we use ai and before i start uh with our company i like to do a few general comments and one of the headlines or actually my favorite headline is is this one from bbc from 2018 where it says that ai will take about 800 million jobs by 2030. so the question is should we all feel bad are we the new evil um bringing um suffering to the world and i tried to dig a little bit into the data and the best proxy i could find was robots obviously powered by software powered by ai and if you take the oecd harmonized unemployment rates but also the data of robots per 10 000 employees you find a very interesting correlation and that is the more

robots you have the more technology you have the lower the unemployment rate so it's exactly the opposite of what headlines try to tell us and the reason of that is in the end quite simple because the more ai the more technology you put the higher will be your productivity and every economist knows if you have higher productivity normally you create also new jobs which which is is amazing now if you can you see me can you see the screens jasper hello hello hello ah michael we can we can see the screen it's only your microphone is scratching at your shirt i think that's the only thing okay sorry cool um so the uh maybe the the important thing is uh is here is what people are only focused is this either human or machine but the the what i think is more important and even more frequent is this human plus machine so where humans complement machines um or we are where ai gives human superpower um and and these hybrid activities are really the

very important one and and the amazing thing is that ai in this case brings like a job promotion to you without really having to work for you because it gives you super powers um and that's exactly at the core of i2x we want to give people super power by augmented communication we built a software layer on top of the brain to help people to have better interactions with their customers in the core where we focused on if you take the whole customer life cycle and customer acquisition funnel lead generation totally optimized then the activity tracking hopefully you enter everything very carefully into your crm erp system etc but in the middle where the customer and the agents are on the phone you have the box of the unknown during calls there are no insights and especially there is no coaching there is no real time feedback and and that's exactly the problem we solve because normally you only have um like very specific calls even if everything is recorded where you dig in and then

after the call you can um kind of see what went well and what didn't went so well but the great thing with i2x is that we do not only record every single call we do not only analyze every single call but also we were giving coaching during every call so in real time how does that work we tap into the call we have like dozens if not over 100 different integration possibilities by now close to every telco system in the world we transcribe it then with our own proprietary speech to text where we have it built a very advanced self-learning ai then this is analyzed um visualized and in real time pushed back to our platform and to the screen of the agent so he can guide and steer that conversation better now what does real time mean in our definition it means less than 0.5 seconds for a full cycle to the cloud and back to the screen because only then you're able to take the results and and kind of put them into

practice during your call as as a high-level picture it looks like on the left side you see the customer the agent interaction the data goes into the platform and then goes back immediately in real time to the agent or it gives you insights um and coaching post call but also it gives insights and recommendations to the team leads to the trainers and so they can take this they can agree they can augment it they can verify it and then again this data point is taken um so it can give better feedback to the agents and this is really it's like constantly self-learning self-supporting circles so with every data point every input the platform the app gets better how does it look like for for the agents they either see a tiny box on their screen which doesn't disturb or they after the calls have the possibility to get into advanced analytics then and if you um kind of enlarge the app you see there's like gamification elements it gives you certain topics or or words

they should say in the right order down to should you speak faster or slower high level the ecosystem we've built we really try to build a 360 degree ecosystem it's not only the app and the agent inside platform it aggregates to the team leads it then aggregates to the platform to the manager of certain locations up to the headquarters of corporations and no matter if you're a small company with 20 agents disbursed over a country or a huge corporation with maybe thousands or tens of thousands of agents it gives the management the c suit the possibility to basically change the course of thousands of people with one click of a button so if you want to push another product have a different pricing point wanna implement some some new marketing or worrying one button you click it and within a second bam goes to thousands of people but not only are we are we focusing on what should be said also we go into complex communication and emotional analytics things like friendliness helpfulness is there

an individualized conversation do do the agent communicate on eye level with the customer and many many more highly emotional topics it furthermore a new product um we we are launching is that we automatically can analyze the customer satisfaction ratio the nps score so instead of sending out emails after the call just by listening to the calls directly after the call we can tell you how happy or unhappy a certain customer was so and and that gives you totally new possibilities um of both analytics but also then during the call of of feedback and of guiding a call results so far are extremely good we have dozens of happy clients and corporations that use i2x in intel core energy um a lot in in insurance retail et cetera and agents are super happy we see 10 sometimes 20 30 percent increase in conversion team leads love the product they can do coaching a lot more efficient with 50 um time saving um and it help really helps them to achieve their kpis managers suddenly can do

true kpi based decisions they can a b test things and they also see a lot less agent attrition which is a huge problem and especially with sales and customer service agents and ultimately what's which is most important customers are also happier because they get have faster calls get faster to the solutions and are served better in these times also very important what about working from home as you can imagine a pure cloud api first solution um is is is is kind of perfect for this this time um people are not left alone in their their homes but they get coaching support in real time like an angel um delivered by uh by i2x and also um with we're very happy and thankful for many different awards and prizes we've won over the last couple of years um which confirm the path we have taken um and that's it thank you very much if you have any questions feel free to reach out for us um we're always looking for obviously new customers for partners um

for people who want to maybe join us um and apart from that obviously very interested in everything related to technology and ai and also a huge um thanks to to fabian and veronica and the whole Rise of AI so thank you very much thank you michael thanks for the presentation um surprisingly i have questions for you um so we are we are based in germany here um trade unions are pretty strong also in europe i can imagine in the us it's a bit easier to just observe people but you're kind of building a big brother here so how does i2x work with trade unions so so first of all um i we're definitely not building a big brother i would rather con compare it to an abs system in your car that kind of steers you and and helps if there is need um that's once secondly we're really not there to kind of observe or to to kind of evaluate people but really to help them we're a coaching system a support system um

and that's that's very important and obviously in nearly every um medium-sized and certainly every larger corporation there are works councils there are trade unions they are actually great partners and um certainly um some of them are very critical at the beginning but so far we could uh convince 100 of them that this is an amazing solution because it's exactly not kind of too like a computer saying oh somebody should should leave a company or not but really to make them better one of the largest compliments uh for me is if if a call center manager or location manager says look there were one or two people um that were not good enough and that weren't making the cut and i was already thinking about letting them go but thanks to i2x we could bring them to a level and coach them um where they actually now are performing so well that they can stay with us long term um and and and this is exactly the core we get like every single call every hour

we get feedback we say think about that do this do this it's like the computer never gets um kind of tired it never gets angry also people appreciate a lot more feedback from computers than they would appreciate it from maybe their manager who after the third or fourth time maybe isn't that friendly anymore um so there are a lot of advantages so i can imagine a lot of people want to call you now and implement i2x but what are i mean you're not you're not ipod yet so what are the use cases you're currently addressing i mean would you would i as a vc could use you in my company or are you rather targeting other groups um well theoretically you could and finally i actually have a few vcs who say like we have no clue how our analysts and the show associates are pitching our franchise so wouldn't it be great to listen um and and to help them to do that better but that's certainly not the core of what we do

so the core is we serve medium-sized and large corporations where you have dozens or hundreds or even up to thousands uh of agents on the phone with the customer um where oftentimes you also see kind of attrition so change in the staff of 20 30 40 per year um where you have to onboard them where you have to train them you have new products new pricing new marketing campaigns um and then especially for in in areas of telecommunication we're strong for oh sorry um for um for for insurances that's a good use case for anything related to e-commerce some actually interesting cases for medicine for example now obviously now there's a lot of talk how do you do kind of medical trials and it's very important that over several weeks months or years you ask every kind of patient or everybody who tests a certain medicine the same question in the same order and you have hundreds or thousands of them you have sometimes hundreds of doctors and how do you make this sure our

solution can make sure that you say that it's like it tracks like and helps people now say this say that say that and it really measures it so you can do medical trials very well these would be some of the use cases where we're really good all right michael thanks a lot i can see a lot of analysts listening right now getting nervous and a lot of partners getting very interested and happy let's see you know there's just one funny thing there was one we see where this was at where one partner wanted this for the other partners that's a very interesting thing don't don't give my partners any ideas please uh all right michael thanks a lot um thank you very much bye-bye bye-bye that was ai and voice and now uh we're moving over to ai and text so we have a high performing entrepreneur coming up now he's a world champion in canoeing he built and scaled an e-commerce company and then he wanted to solve his own problems by himself hence

he founded the company solvemade eric von miller welcome thank you jasper for that uh introduction all right thanks everyone so i'm happy to be here as you said after uh michael bream uh it's a pleasure for me so i'm the founder of solvemate we are an omnichannel platform for automating customer service enabling meaningful conversations between end consumers and customers so a quick agenda um quick intro why this is important now how we do it and then because probably we've heard a lot about this topic how we are different so let's get started with an introduction um my name is eric as i said already and thanks to the introduction from yasba in my first life i have been a professional sportsman so if you know the olympic sport of whitewater canuslam with green and red poles hanging there i've trained for more than 10 years and ultimately led the world ranking and became a world champion then decided to quit and now since more than 10 years um founding and scaling high-tech companies on a

private side i'm a father of three kids and enjoy hiking and mountain biking in nature and i'm a strong believer in the ghoul in the golden rule of treat people the way you want to be treated so here a quick intro about solvemade founded his company in 2015 we are a customer a software for customer service teams i will explain in a second what this is about it's quick and easy to use not for developers for business people and our focus is to smb and sme small to medium sized businesses and enterprise with 10 to 200 service agents our clients are happy that's very clear we are solving millions of conversations each year and we are very culture driven so you will find online solvery.com culture our culture manifesto it's a 15-page constitution how we interact with us and clients and every stakeholder ultimately we're very international team we are from 20 different nationalities now you're curious what soul mate is so here it is we are a b2b sas company we leverage machine learning

our chatbots act as the first line of defense so for example if you look on this small animation you all have seen chat bots which is the top of the iceberg after answering a few questions you come to a solution let's watch it again you answer one two maybe three or four questions and then you will get shown a solution a solution is what you would get if you would call someone on the phone in customer service uh this is kind of what it is about in customer service automation to get to solutions um in the second example we have just seen a real-time database request to fetch personalized information that you want to have um in customer service so we just talked about ai and i'm very often asked what are you doing there and so i thought i'd make a top-down view on what this is about so yes we are conversation ai conversation is the back and forth uh in text with um the end consumer ai because we leverage algorithms um

and then there is use cases and we've just heard about another company i2x that is um also doing conversational customer service they are helping people on the phone and they are doing something else than we do because they are in this agent augmentation thing so they are helping people um getting better we are deflecting so it's about making sure that people can help themselves in self-service and we are not a tool that helps the customer service agent so why now um we live in a world where there is billions of customer service requests every year and i know that every one of you has been on hold on the line and we live in a world of abundance time is the new goal the only thing that i have not to waste is being on hold in any line or waiting for an email to come back what i want is 2 a.m at night on sunday morning the same customer service that i would get if i would have someone on the phone in

real time but companies can't afford this to have that 24 7 perfect service response times so we live in an instant economy i want to have netflix and amazon everything shall be instantial why shall customer service not be instant some chart um that you know from our good friends from gartner you might have seen the hype cycle for artificial intelligence and you can't see it as really small so i added some arrows that is chatbots in 2020. and let's look at the development over time because gartner is doing this year by year it started in 2017 2018 2019 was the hype cycle of chatbots according to gartner and guess what 2021 will be the year of the bots and we have been doing this for years now but i strongly believe that the majority of customer service leader will understand that chatbots are no fake anymore they are through the hype and there is companies that deliver true value so i'm excited for the future and um i quickly want to talk about our approach

so people often talk about chatbots but in customer service automation it's not about the chatbot it's about much much more and the chatbot is just the tip of the iceberg it's about user authentication it's about having and a way to know that this user is authenticated as he is in the app or logged in into the interface and what you can do with that you can connect to a database which is everything built into our cohesive platform that you can personalize the experience and i don't mean say hello eric i mean fetching real-time information pre-filling forms and triggering processes directly on the database because that's what self-service is about for the most requests i don't need a human i just need to say what i want and then i want to execute also sometimes humans are needed and that's important so if i want to have an agent to process my request we are routing to the right agent this is not automation this is routing efficiencies because the agent get all the information all

the context and i have the right person on the phone finally an agent is always a click away so agent escalation is of course inbuilt which leads to a lot of benefits in addition because we are so numbers driven where we get real-time insights we of course have a c-set measurement ultimately we force our clients to think about processes because customer service automation is about process definition and leads to product innovation because once you know what your clients want why don't you feed that back into the product and find out uh how you can prevent customer service because the best customer service is no customer service and yes it leads to cost savings but that's a secondary or tertiary benefit how do we typically work with clients just a very quick note uh we are a sas tool and we are built for customer service teams um we do a kick-off workshop we train and configure the bot the fastest client went live on the same day but it typically takes a few days of

work and a few weeks of time we go live we optimize and change the flow of the bot and ultimately typically clients understand that they need to route more people to the bot in order to have more of the great customer experience and clients benefit from of course 24 7 support we guide clients along the way and and they're very fast go live it's not a month's project it's something that can be done in days and yes of course we have great numbers to show if you look on our website there is case studies where clients benefit um from deflection and routing and you might see well i've seen a lot of those numbers and people are talking about this and yes it's right so doesn't every vendor promise this probably because it's marketing numbers but the point is we are different you might laugh we are fundamentally different in how we approach this technology and i only have two more minutes left in this talk so i only want to scratch the surface of

what we did we've invented a new way of creating a chat flow let me repeat that we've invented a new way of creating the chat flow to find the client's intent we call it dynamic decision trees and it is a logical framework of questions and solutions that are fed into a system by humans by experts in in the of our clients and out of that we are dynamically creating a chat flow that is that doesn't use nlp at all and don't get me wrong nlp is awesome but the nlp is added on top without additional nlp training i cannot go into the super details here but you might ask the question why has nobody else taken that approach yet if this is so good eric and the answer is i got a call back in 2015 where i was researching about machine learning algorithms and that call was from my father and i was his customer service agent and he told me son i can't send emails anymore please help me so i was doing

diagnosis questions like it is in customer service you want to find the one out of the 100 intense fast and i asked a series of multiple choice questions and i figured out that nlp is great but there is a big hype around deep learning and lp and don't get me wrong again i'm dictating whatsapp messages every day i'm using siri every day but for customer service automation it's not the tool to use and we just took a different route and it proves to be a very good one as a last takeaway from this uh call um from this presentation a quick summary of hopefully that you can take with it there is different ways of approaching a chat flow so you can have static logic trees everyone knows you drag and drop a chat flow in a browser interface it's hard to maintain for the trainer and because it is visually there um the effects if the bot trainer leaves the company or this thing gets super big it's a challenge because there's no algorithm

supported if you do it well the end user has an awesome experience if you look on nlu it's hard to maintain if you've got a lot of intents and utterances in dialects and languages and we all have had chatbots where you type something in and you get the failure state but there is no limitation to the chat flow of the human because an algorithm is supporting ultimately you could have guessed that we are green in all these things it's super fast to set up due to our special algorithm in the setup and the training we can have as many solutions and a tense as you want and it's as good um as other interfaces and with that i am done and say thank you for listening thanks a lot eric thank you um let's start with some questions again please um i can imagine you were talking about deflection and actually taking over the work of customer care agents now before we heard about augmenting this seems to be kind of replacing is that's what

you're doing or also merely also augmenting you asked the same question to michael braim and it's interesting because the positive interpretation is i i've had customer service agent telling me you know i was tired of answering 50th time the day how can i reset my password and giving all the status updates i am relieved of the robotic work and i can do the true empathetic work and ultimately that is increasing the skill set of a customer service agent which will there be less customer service agent 10 years from now probably but the ones that are there will very clearly have a higher skill set and be more well-paid uh jobs and i can imagine that it's it feels sometimes vcs should also use your tool um i would i would say for starting this you probably are a bit more careful as a new customer so what are typically can you explain to us what customers are doing at the beginning what are typical cases they're using for and what are the typical customers that

really profit benefit from your from your tool game so we typically say in order to benefit from automation you should have minimum 10 to 15 customer service agents because don't underestimate the bot it is some work to maintain we say a few days of setup and then maybe half a day per week to maintain so if you have less than 10 agents you can make the math that you're adding work for yourself you need to pay license fees because we're a software vendor and um so you need to have thousands of requests in a month to make that work and then we are industry agnostic so ultimately um every industry can benefit from automation talking about ai you already mentioned it a little bit so what type of ai are you really using in production and maybe also giving us a little bit of outlook because you said 21 is the year of the bots what will happen with ai and bot that's a very good question we got four different algorithms in our product

we got the mathematical flow generation algorithm which is kind of a bayesian probabilistic system we turned this algorithm around for training the system it's kind of the inverse algorithm we have um nlp as addition on top of the ddt algorithm in the chat flow and we have an nlp clustering algorithm if you're dealing with thousands of conversations to group them for training what we're going to add in 2021 is we're adding voice as a channel because voice is remaining and our bots with our structure are are determined to go into voice so that's going to launch in q1 next year and we will add no code flows because customer service teams want to own processes and model them so we are connecting to the database and then you can drag and drop if client does this do this to execute processes which basically gives the customer service team even more independence from it and from the scars resources that they don't have access to cool that sounds very promising so in case you're interested please

contact eric we have the website in the chat thanks very much eric for being here thanks all right coming up next um is jochen deisser jorgen is a lawyer by training we're not really used to this physical thing here it's a lawyer by training he's seen both sides i think clifford chance but also amazon um and he is now an entrepreneur and he will tell us more about privacy solutions welcome john thank you yeah welcome i'm joking i'm uh i'm i sometimes feel like an artist trapped in the body of a lawyer and hopefully with ai for privacy we are taking data protection to the next level but let me tell you where it all began here's back in 2017 i was sitting in my old office with my lawyer hat on with a number of colleagues experienced data professional data protection experts business professionals and we looked in amazement at the numbers in front of us this team of experts had just spent 70 hours combined on creating one data protection impact assessment and

for those of you in the know that's whatever you have to deep dive into the data protection regulation for instance whenever you launch an aii solution in your company and then at that point the regulator had just lowered the threshold for uh this data protection impact impact assessment through the coming of the gdpr the general data protection regulation that we were suddenly not just looking at 1 10 20 dpias we're looking at that one customer that one client hundreds if not thousands of dpias that we had to do and when we look back at the at the gdpr what we think of are these tremendous fines that can you know snap the existence of a company within seconds but we're also looking at a lot of formalistic pieces documentation risk assessment process changes all that translates in a lot of costs so we had to find another way and we also had to find another way because as a lawyer i could have just sat back and said hooray i've got this good clock here

let the times roll and the cash flow in but that would make it very prone to manual risk i would sometimes other people would miss important aspects we would have inconsistencies in the approach of solving these issues so that's when we expedited our idea of creating a data protection software the privacy suite because data protection can be achieved so much more efficiently with software and the privacy suite was ready for the gdpr it collects records of processing your data processing activities it reviews them assesses them on a rules-based system and already then atomized the time and effort required to do for instance at dpia but also to comply with all of the other requirements under the gdpr it's life worldwide it covers thousands of records of processing at the moment it's used by a number of customers and they have expedited their their way of managing data but it's only them they've created their own rules the knowledge that's contained in the privacy suite is proprietary to those customers so we thought of something like

what if we could combine the knowledge in the privacy suite and make it available to everybody else if there's a an enterprise that uses a certain ai technology and a new startup with no data protection knowledge whatsoever comes around the corner how can they too benefit from each other obviously sharing that kind of sensitive data is is an absolute no goal so we had to come up with the smarter way because we believe the smart use of a.i needs to benefit society as a whole needs to benefit more than just one company and then comes ai for privacy the idea behind this is that we take a case-based reasoning methodology so we look at all of the existing processing activities and analyze them to create a case-based and if a new processing activity comes in that is then compared with the case base and the most suitable the most the best comparable processing activity is sought and found presented to the user who can then say yeah that that works or it doesn't work and

um the the user interaction is then captured and put back into the case base so that the the case base continuously grows irrespective of whether you're a startup or a multinational enterprise and we're very proud to have been granted an innovation grant by the german state of hesse for this project but how does this look in uh in practice let's look at two scenarios first of all we're looking at the screening we're seeking we're trying to detect the risks inherent to a processing activity so already the privacy suite identifies certain risks based on the rules with um ai for privacy there's a new there's now a new recommendation a suggestion there could be another risk here or there and the user can then um uh review that and either accept it or deny it and then the risk detection is already improved but secondly and that's where it becomes a lot more important a lot more thrilling even um is the actual assessment of the risks and not just the assessment of the risk because

lawyers like risk air uh but from a business perspective we want to reduce the risk we want to do something about them we want to we want to implement the solution and if that means we have to overcome the risk well then let's go that mile and that's where ai for privacy can really do make a tremendous difference we believe so again we have a processing activity candidate management with automated analysis of the application and our rules engine already comes up with a risk score of 5.5 in this scenario then the ai for privacy engine looks at the processing activity and suggests additional what we call mitigation measures to further reduce the risk and the user can then take a look at these risks review them and choose whether or not to accept them to implement them and thus uh further reduce the risk and the cool thing is that any kind of user will benefit from the whole global know-how of all of the processes that have implemented um risk mitigation measures for this

processing activity for this ai tool somebody smart might have thought of something really cool and it's now available through the engine for all of the users of the privacy suite so that ultimately the risk would be reduced to 2.3 in this case but we believe we're just at the beginning well i've worked for amazon so we're day one of data protection in in a way ai has tremendous potential for data protection we're just at the beginning and and here's my here's my one other thing for the role uh because instead of seeing data protection as a blocker for innovation as many companies perceive it and we've talked a lot about regulation at this conference as well instead of blindly pouring concrete into innovative systems by following what the the state regulators think is right in their two three years behind attitude or what uh data protection officers who cannot be terminated in the company consider as the lowest risk approach i think we need to disrupt data protection we need to disrupt it in a

smart way in a in a balanced way and approach it from it from a different perspective and embrace it uh so to speak because technology will change we've seen super interesting solutions and when we look at fabian's uh keynote in two three years they'll be on the oh did you remember that well it didn't quite work or it's been working and it's been replaced by something else technology evolved so quickly so success in the long term is when you treat people with respect when you treat the personal data in a data-driven world with respect and that's what we think is going to be the key for long-term success so make privacy happen and make it happen not just because you have to but because you want to because it does matter thanks thanks a lot jorgen yeah that sounds like a very very relevant topic um maybe maybe starting off a bit easier um how do you how do you start with your product it sounds like you have a lot of regulation that you

have to read that you have to implement that you have to put into some kind of software kit um is it is it people based is it ai based yeah that's that's an excellent point jasmine that's one of the when we look in the glass bowl where would we improve the software the intake at the moment is completely manual i mean we can read databases and and match uh the data fields to data elements to data categories but that's that's that is very it's a very manual process and i know there's a a number of companies out there who try to automate this this intake process and that's going to be the the next big thing if we look at the whole food chain of data protection to to automate the intake then automate the data protection assessment what what our core technology does and then ultimately also automate so what you know the implementation what's going on with it but at least you have you have the data already in there that's i guess

that's an edge um how how would you approach than a new customer um probably everybody has a different ai solution it's customized people do their their things do you have a general solution for them or would you first go in talk to them adjust your tool according to what they're doing my customers are lawyers or data protection officers so they work with excel and for them ai is if then and if they're advanta advantageous else uh so we're starting at a very very low level typically uh if if we're lucky they have their data in in this in in one central database so that we can implement it and input it in our tool what we usually find is data is completely decentralized and um we have several ways of putting the data into the tool but we see the privacy suite as the one-stop solution for data protection at any given customer so not only do they do the risk assessment there but they would also manage the data protection agreements they would cover

access requests by by individuals but it all depends on putting in the right data or getting the right data into the tool and at the moment that's that's a that's a very manual process and um ai for many is in that community is is a threat rather than something cool with a question from the audience from jochen i don't know you want to ask it yourself no but are your customers ready for aren't they just moving away from excel sheets to software yeah it's a that's that's a that's a big challenge uh so we have a a lot of convincing to um to do um a lot of um uphill uh battles but ultimately uh with the the more and more data-driven world there's just no way that people can continue managing it in excel sheets or even within like physical uh files anymore so uh we have great customers uh on board we have uh uh very innovative companies and and those are the early adopters in in this world um if we come

to the normal customer with our ai solution uh well we might have lunch but that's it now maybe one last question um when you look at the space and all these people now playing with ai working with ai earning money with ai what are the typical easy to cover mistakes people are doing and what are the things where you would say you definitely need a lawyer that's not as easy but is that something we can all avoid that many people do yeah i think we need to change the mindset a bit so we need to um at the moment we all see privacy and ai as like two opposite topics and and and they they they're like trains running against each other uh because what i feel a lot of times people see a black and white on both sides like that's what i said the data protection officers or lawyers in the companies uh they're so risk-averse whereas on the other side with um the folks in the eye community uh are you know

very entrepreneurial very very um innovative so this mindset needs to change and these things need to be combined we had in a talk earlier today the the double phd problem that we need to combine these sort of controversial disciplines in order to make it to make it happen because if you're creative if you if you if both sides move to each other um there are a lot of solutions and uh it's not as as dark uh the world's not as dark when data protection comes around the corner you just need to be a little bit more creative and maybe a little bit more risk friendly also on the lawyer side that we have people reading the stuff and tools make it easier thanks so much for being here yeah my pleasure jasper thanks for inviting me thank you thank you all right um next coming up still on stage um i'll just click here i guess um is clemens vienneckel he is from the mirantix uh incubator he studied at sagan university and he is

truly visionary in terms of hyperscale ai so he wants to help us to manage unstructured data with unsupervised learning please welcome clements hi thanks for that introduction uh jasper so um as you said like i'm one of the alumni software neurotic exercise ecosystem i'm happy to be here thanks for organizing this great event and i'm happy today to talk a bit about our work at sia search we are a young software company um now about 10 people in berlin but let me actually start with the problem that probably a lot of you guys know is um when you when talking about unstructured data management you probably are aware of this problem from sorting your holiday pictures when you only have time dimension available in order to find data then it can take hours and hours in order to find the relevant stuff and we essentially address this problem like in the industrial setting where the problem is like the same but in a much much larger scale so companies who built really want to build

really robust computer vision like instead of driving start to create data systems with billions and billions of frames to train their to train their models and working with this data still remains a big hassle what do we mean by that when we look at structured data you typically have like a whole bunch of nice data management systems elasticsearch postgres and all of these databases and you can just like query the data retrieve whatever you need understand what's in it you can even check if it's biased or not build a dashboard maybe even do some nice machine learning on the unstructured data side what we see here say recording from a car there's no such thing so in order to work with that kind of data you essentially have to work with the raw files um and with the actual like direct direct recordings and what are these what are the engineers typically interested here when they work with this data one they want to understand the content of the data so what's actually in it

to know if if and for what they can use it to train their models or two if they have understand that understood that they want to retrieve very specific sequences of that data in order to then forward this to their training or their labeling workflow like this is really a big problem the answer is yes if you talk to companies working on like industrial computer vision it turns out that up to three quarters of engineering time is actually spent on so-called data handling activities which is like all the activities taking data from raw data from storage and eventually using it in the model and this takes so much time because and i will go into the workflow in a second because the interaction with the data is so is so manual so what csu does essentially supercharging computer vision development by introducing a new way of accessing raw data what does it mean we build software that runs on the customer's data lake and ingests raw data and essentially builds on top of that data

lake a very very lean layer of metadata that then can be accessed by the engineers via award-winning graphic interface or api what the engineers do with the data they then essentially can just use an ap use the use to metadata to query aggregate understand slice and dice the data for annotation training and evaluation of their models how does that metadata is how is the metadata generated there's multiple ways to do it one you can run algorithms that detect certain events in the data of which we have a couple already pre-trained for specific use cases you can also use metadata that already exists most of their customers have inferences from their models have manually created metadata but they're still not able to search for that metadata on all of the stuff that they have that they have recorded and see a search center enables to ingest it that other metadata and add it to to your to your search um we have like unfortunately this should be a video uh but like yeah so we have

we have a nice api that like and gui that enables you just query directly the sequences that you're interested in and you will get like these exact frames so let's take a quick look at um the workflow of how it looks today so to understand better what we're doing so typically you have data acquisition then you ingest the data and most engineering time actually goes into the data handling then forwarding some data to annotation and only like a minority of the time being spent on actual algorithm development and training and validation with csh we try to accelerate that cycle by at least an order of magnitude and increase engineering productivity by essentially adding automatic generated metadata and then using csh tooling to speed up all of these search aggregation and selection tasks that now take up so much of the engineer's time hence expanding the time that they can spend on the algorithms and hence overall accelerating this development cycle why are companies not doing this today oftentimes a lot of the companies that we're

working with have some metadata often basic but they're still not able to search for it on scale or they don't they don't start they haven't started creating more metadata the reason is that if you have billions and billions of frames adding metadata can create a big mess because you're storing a lot of similar and redundant information and have a lot of emptiness for all of the other frames that don't have metadata and putting this into existing common database systems turns out to be like very very slow and efficient so what we've come up with when we looked at this problem is we essentially introduced a new data model for the metadata that we put on top so when you think about an event that's happening in a video stream like in a car here a lane change instead of stabbing metadata for every single time frame we use an algorithm that only detects start and end point of that metadata of that event and essentially store that in in our metadata base as a result

we we have implemented a new way of accessing the data through this this uh interval match to an interval matching logic what does it mean in a classical database which you see here below you would have metadata for every single frame and a lot of redundancy and emptiness and then when you retrieve data it takes a lot of time it's incredibly slow if you have like thousands or thousands and hours of data even just on like 100 hours of data and can take many many minutes to retrieve specific sequences so we've developed like a model that essentially uses only these start endpoints and inter and matches the intervals where they intersect which is like not only incredibly much more efficient in storing the metadata because it's just like much less and doesn't require any redundancy it's also incredibly more efficient to retrieve it and when i say more efficient we're talking about three orders of magnitude which can be really meaningful if you have like many many hours of data that you're looking at so

we're already working with like a lot of we're only like about a year and a half old but we already work with a lot of companies um we have like run a series of pilot projects or we have like a smaller license integration going with the first customer and we have like many many companies who are really excited about the product already testing it and are are like um having the engineers on our platform to try it out um we have also an open source version i have in qr code for that later um which is essentially like you can try out see a search for free if for for academic use if you know of new scenes a very famous public data set uh you can sign up on our website and just look at look at this data set through the lens of like accessing it on a semantic level we have seen a lot of great interest for that a couple of hundred sign ups in the in the last couple of

weeks uh because even for just a few hours of data like new scenes it's incredibly useful to just search for many many semantic dimensions um and get the exact time frames just in an instant but this is by no means limited to autonomous driving and robotics where we mostly started we now have like first prototypes running for other computer vision use cases mostly working also with smaller players here in berlin ranging from retail checkout to drones to plant inspection and even human gestures because the problem that we're solving is really just retrieving sequences that are associated with like semantic semantic dimensions and this is like a problem that probably everyone working on a larger data set for computer vision has already encountered as a next step we're looking towards also really establishing csh as a middle layer between data users and data producers for that we're starting to establish a marketplace for data annotation transaction where we have a couple of leading players in the field already already sign up who like want to participate

in this in this workflow um so they essentially like can look at a place where customers just put out data that they want to get annotated and already know what's in it because we have produced metadata about it and then can get offers from leading annotation providers with a single click from our product in 10 years we really envision cs search to become the standard gateway for for sensor data and to enable a whole range of computer vision applications that will affect many industries and we have built a very very strong team um from uh from top-tier institutions and and companies my co-founder mark has done award-winning research particularly on data driven development at edh and zurich um and we uh if you are interested to learn more about what we're doing like feel free to scan this qr code you can sign up for a demo version of our product um and uh and look whether this might be interesting to you if you catch yourself uh looking at file to file data in

your development uh we definitely should talk thanks thank you that was way ahead of time i hope everybody was following i was following i was really excited thank you very much what i understood so far is um you're basically helping everyone that is working in the field of computer vision um but to start with there is a certain metadata required so if i come up with any new random thing that you haven't been working before you would have to set up that first so everybody who wants to jump on this product now yeah okay so what are the areas that you can serve right now specifically so um yeah so mostly most of our work has been focused really in kind of like different sized robots which include self-driving cars so there we have already essentially like a series of models who detect certain metadata in the data automatically but in the in the midterm and i think thanks also for asking that question most of the time the customers are actually like more excited

of also adding whatever models they have to the system because they typically produce already a lot of metadata either they add it manually or they have their models running and they want to get have this metadata and connect it to everything that they have recorded in a meaningful way and there's just like no solutions out there to do this which uh then um even for ground truth data so they get back labels from from the companies that do them manually and even then this just sits in files and they have no means of relating this to their raw data so actually i want to enable customers to build this layer of metadata in the most meaningful way for them possible okay i understood and you said it's 75 percent of the time of a data scientist what what are the data scientists experiencing right now that are using your tool that sounds like they must be a bit like what michael and and eric described they must be totally relieved now yeah like ideally i

ideally for for for the for the first like try users that are testing this out we want to replace as many kind of uh data interaction activities that they do now essentially with the with the raw files with essentially just uh running a command in the api or query um like with a certain metadata associated to it and hence essentially avoiding people to go to the server and download the batch but instead just like working entirely on this metadata layer so our core kpis you really like how much engineering time we're really replacing and for that like the the core dimension is in terms of like what so we work there really closely with the customers as well in terms of like what are your engineers doing today like there's a lot of overlap there's also quite some differences between companies uh what data are you looking for and why typically and essentially figuring this out and then seeing how they would like implement these workflows uh with with our tool since you addressed the

open source community what what are you hoping to gain from them what would be your expectation but also you ask maybe to them so sorry i have to like do a quick correction here so um so the the version that we have released for free is not open sourced yet at least uh so so not that i i i cited wrongly here uh we are thinking about open sourcing part of the technology stack as well um the version that we have out now is free free to use we are hoping to establish this new way of really storing metadata about sequential data um as a standard which is also why we are very um actively looking into um yeah making essentially a core part of this data infrastructure open source uh and uh yeah stay tuned for uh to come 2021 on this cool thanks a lot clemens thanks for that yes all right we're now moving over to another virtual guess i think dialing in from the beautiful city of leipzig christian pisnik from

vandelbots he recently raised 26 million euros if i'm correct from very reputable fund from israel and london 83 north and microsoft and others and he actually has robots working so this is a nice move over from the previous topic so very glad to have you here welcome so hi everyone my name is christian pisnik i'm a ceo and co-founder of of a startup we are based in dresden called bundleboards and basically we are active in the field of industrial robotics and we are providing a technology which basically enables everyone regardless of their technical background regardless of their education regardless of their cultural background to actually program industry robots and i was i was asked if i can talk a little bit about the about also some backgrounds and challenges in the robotic market and this is extremely interesting because if you if you look at the market of industrial robots everyone thinks like and you can read this everywhere that robots are capturing the world they make us all unemployed and um it's it's uh

it's a fact that the robot market in the past approximately doubled um every five years so the market um is growing continuously however if you look at actually how many robots are currently deployed globally compared to human workers you will actually find that less than one percent actually of all jobs globally in manufacturing which could be done by robots really in manufacturing are actually done by robots so there is a lot of potential to grow and actually we are active in a field where the perceived advancements in technology and the reality is extremely different the problem is today basically every robot manufacturer every robot um a provider provides its own proprietary technology stack and we're actually talking about technology stacks programming languages that directly comes from the 80s so one interesting example would be denzel or one of the largest robot makers they have decided in 2019 last year to introduce basic as their programming language in order to to to basically make their robots do something and for the end customer currently basically the

manufacturing companies and the vast majority still today is automotive or electronic companies um it deploying robots is rather expensive so you spend in average 75 of the overall total cost of ownership comes with programming robots and then those jobs can only be done by very specialized expert so yeah those are usually employed by external companies called system integrators and this makes it super complex and you can barely reuse any any code of um you have deployed on one robot and finally they are all rather disconnected and our company basically has worked over the past years on a technology i would like to present you right now any given mobile applications the revolutionary and stop this progression next is um production systems me so um to to basically sum that up very quickly because we have we don't have too much time right so we have a technology where you basically can show a robot that has by example using an input device which is um tracked in in six dimensions plus a little bit more

of metadata such as acceleration and based on the demonstration um our software is able to create a model of the application itself so it's not a one-to-one repetition it's an abstraction a model generation based on data using many many different methods coming from mathematical optimization up to up to really machine learning so deep neural networks and and this gets done gets better of course with more examples but in already right now the the first sketches are are really really good and then you have a very easy to use user front and basically to to adjust and fine-tune what was generated by the software which which then later on is again part of the feedback loop so so with uh with with every single installation and every fine tuning basically it's get it gets better and then um and then what you what you can do at the end is generate code for specific robot platforms so you can generate code for kuka abb uh universal robots whatsoever in the host language of of the of

the individual robot so what we can do with that is basically go let's say typical gluing applications grinding welding basically reduce the time to programming up to 70 times faster than with conventional text based programming methods and basically enable non-programmers so non-trained personal who are task experts to really do it themselves and then besides that um on the left side what you can see is basically the fact on how connected robots are today so there are systems which are which are called plcs so those are basically um systems managing more complex systems entire cells or are part of the production line and the robot has an interface to that so communication is always point by point very individual code what we can do is is generate the code that automatically connects robots and we are very closely working for example with volkswagen amazon on what is called the digital production platform as well as microsoft on azure and the open manufacturing platform to and and this basically um has an automatic gateway to capture the

data of robots in production in the first place to enable more advanced intelligent robotics in the future the typical companies we work with very quickly are come from not only automotive but a big portion as well as pharmaceutical as well as semiconductors electronics metal food whatsoever and then our technology is applicable for for many many types of customers so from from a small bakery with 17 employees in girl it's up to large tier automotives such as volkswagen and one of the coolest presentations we had in the past was with microsoft where we basically presented our entire technology stack to the microsoft board and satya nadella the ceo of microsoft was able to teach the robot a pvc sealing process over a distance of 8 000 kilometers so he was using the trace pen in redmond the robot was in germany and he was able to process stability program that in a very short amount of time and and i would like to finish my my my talk with maybe one very quick example from a

volkswagen uh production facility where you really can see the power of the technology plus ai in its whole so this is a video in the what is called end-of-line inspection and what you can and basically what people need to do is like check whether a car was assembled correctly especially now where every single configuration of the car has to has a carbon dioxide check and what you what you need what you need to what you need to check is whether for example the liquid in a can is correct whether the correct labels are glued and this is today entirely done manually and in order to do this adaptively with a vision based system you have more than 800 positions you would like to check on such a car so what you can do is just point at the object you would like to check with the trace pin and and the entire robot control is generated with just one button press and the robot is taking pictures transferring this to a back-end system in the

cloud environment where you can label those pictures and actually define what parts of the picture define whether whether a part was assembled correctly or not and um yes i'm running a little bit over but this video is very quickly over um and this this basically trains a neural network and because um the pictures are always taken from the same angle those are always the same objects with a similar lighting it only takes 20 to 50 reference pictures in most cases to get a 99 of accuracy and the coolest thing is that you can share the both the learnings from the from the robot so the robot control as well as the image processing across all of your factories globally and this is just one example so thank you very much and and eager to hear your questions thank you very much totally fine you ran over time i put you in the wrong city and coming from bremen it would be like somebody put me into hamburg so sorry for that um so question question

regarding your use cases um it looks like you're expanding the market for for robot usage because you put it giving the last example you put it in a field where human were in there but then also you said something about faster training so what's what's the the field you're most active in so so convent traditionally speaking of course the the let's say um advanced and mature automation companies especially in the automotive sector so also companies like volkswagen bmw whatsoever but there are many many companies who now step into robotics also large companies especially chemistry and pharmacies is a big part where robots are not very common today which are now stepping into this game and then there are many small and medium-sized enterprises who look into robotics for the last five six years but what's missing basically is a technology that that enables them to to do adjustments very quickly to get really get access to the technology in the first place and what i really liked was you can do remote training of robots

is it also remote maintenance um or is it is that the key thing um you can do it remotely currently i think all of our installations it's done on site right um but but um and of course you you would need a more advanced it infrastructure into in inside shop floors in order to facilitate that we are part of the 5g lab germany and very heavily also working working with low latency and 5g campus network solutions which in the future will open up those use cases there are a few but they will come in the future especially now in those times of covet where you would rather do things remotely yeah you that that was what i was referring to that sounds highly promising also with the with the edge technology in 5g um when when you look into the future of robotics um are you actually the ones enabling innovation or are there other fields complementary to you what's happening there in robotics um yes it's a very broad question there are many things

happening in robotics sadly really innovations on the software side of things um are very rare right so there is a lot of going on especially when we look into the ai sector in terms of perception um what is called for example bin picking where you where you enable a robot to pick arbitrary pieces out of boxes right but that's only one portion of the of entire automation solution so what to do with those things uh how to interact with the environment and so on and so forth so we are still far away where you really really could speak of cognitive robotics where where you can where robots make good decisions and also many there are so many legal aspects which are currently hindering really deployments of those solutions and that i that i got a little more a bit more pessimistic when it comes to timelines but the the the path is really clear right in in the next at least 10 years we we will see a vast majority of robots who don't necessarily

need to be programmed but but using using um different technologies uh coming from the sector of machine learning in order to uh to become more autonomous cool that sounds like very exciting times ahead thanks so much christian for dialing in thank you yes thank you and have a great rest of the event thanks for having me all right next one and last one but obviously not least one coming up silas dalvik from cognigy so we're going back into customer care enterprise level and he will tell us more how cockney is helping companies to make this whole experience so much better please give him a warm welcome all right let's kick it off very warm welcome and good evening from my side here from dusseldorf yeah my name is silas or stylist i'm responsible for partner management in emea cogniji is actually a german company headquartered in dusseldorf that's where all our developers are but we do have a global footprint so we also have offices in san francisco sydney and we also cover apac with

offices and sold in tokyo because we do have global clients today i will speak about how conversational ai will improve your customer service experience and in order to do that let's just take one step back and understand the term conversation ai we heard it before and you're all experts in ai but conversation ai still is a bit fuzzy to some people in the end it's a technology that enables machines or back-ends to interact with human beings for a natural language and language as we're honest voice is the one true superpower we as humans have how we differentiate and yeah if you think about it um that's the idea of my childhood i'm not that old but still he was my idol he was actually able to do that so michael could just talk talk to his watch talk to his car and here tell a command and then an action was triggered and the software behind to clearly understand what does michael want and what do i have to do and to bring that to

life um that is not easy and there wasn't easy so scientists and business people tried for years and it took a long it took a long time so don't be afraid it's my screen on yours that's the first view of ms-dos just to give you an idea of user how user interfaces were looking like back in the days and at that time if you wouldn't know the exact command line you wouldn't get anything done and even if you would use a command line you wouldn't get done much it slightly improved um when apple and windows came up with a visual interface which you could click which was a bit more embellished good looking it got better and better with windows 95 98 i bet that a lot of people bet on xp and you can dub whether that was progress um but it it introduced dynamic tiles and movement and updates and last but not least apple completely changed the game was taking user interface to a completely new level with gestures but at some

point you cannot make an icon nicer or more intuitive the next step is just voice and voice really is the next front end and the core dna of cognitive and basically our founding reason so um we use voice to automate customer service for leading enterprises and our key pitch or why companies are buying us it's very simple we enable them with outstanding customer service at lower costs on the one hand side and it's managed by them so our platform is really about enabling business users give them the power and decrease the dependency on developers you can of course still fully code on our platform but the main objective is to fully enable business users out there and how do we do that it's very simple it's one of the key reasons why our customers like us um we have a low code no code graphical editor so literally everyone can use it to build a virtual agent it's within seconds you can drag it you can drop it um you can just turn in third

things and it ships with 22 languages for now and we're adding a new languages constantly and we also have um pre-built integrations and that's a really powerful toolset because we have standard integrations to facebook messenger to google to sap to salesforce so any backend system where the data is stored we can securely connect it and you as a business user as a team leader of a call center or a customer service team because you guys have some knowledge um what do customers say what do they want why do they call you can build your own meaningful bots within seconds in any specific channel and that is the true power of the tool to really enable you to make it easy with drag and drop features in reality how does it look like on a specific customer case we don't really differentiate whether it's text or voice even though voice is way more powerful um we have connected more than 30 channels so these are just some examples um of the channels we have connected because

they're the most convenient one they're the most convenient ones so for example whatsapp is used by almost 60 million germans uh similar for facebook messenger or whether it's more business focused um or slack or microsoft teams just to give you an idea and these are pre-built so if you become a cognitive client you got them from day one and on the other side when it comes to voice we can hook into any zip line genesis cisco avaya so these are technology partners of ours so we can fully integrate into a call center and then the logic you built in our software as a business users can then be outplayed in a specific channel meaning you build the info once the authentication and so on because it's of course important and then it's channel specific output so if you have a request about your build on whatsapp we can use the power features whatsapp offers so like widgets galleries buttons and you can even send the the invoice um after authentication or the number um as

a pdf via voice that of course doesn't make any sense in that case we would then just have the output via what is the invoice amount when is it you had it been paid so it's channel specific but the logic you build you build it once and you save an enormous amount of time um i've seen before some people are always keen on numbers and here are some across industry across clients so roughly we can say um three out of ten calls can fully be deflected 30 and we also see a remarkable increase in agent productivity and that is an important point because cognitive is really about empowering agents and relieving them because it's a scarce resource they have to do a lot of topics at the same time and we will really want to empower them because we didn't fly as a human mankind to the moon 55 years ago that a human being on the phone line has to reset a password today is that something a bot can do within a second

on any channel at any time on any channel which is convenient for um well for a customer in the end and when it comes to that we are fully agnostic so we just integrate what's convenient for our clients or respectively for the clients of our customers talking about customers we clearly address enterprises these are some references um here but we have more clients than that one very easy example is lufthansa they really face um a challenging year um i don't have to explain it i believe but they are getting overwhelmed with requests what about my flight what about security regulations in italy can i take hand luggage to the flight what about my child i'm a father myself so does my child have to wear a mask and all these questions in various languages at various time points on various channels want to be answered in real time and that is something we enable lufthansa to do and also other clients obviously i work in sales so i love cognigy and i love working at

cognigy um but it's not just me speaking nicely um about cognigy it's also garner we've been awarded a couple of times we are a cool vendor and we also have some nice reviews from our clients which are very happy with us and also when it comes to our nlu so we did the key algorithm behind our software um we outperform ibm microsoft and google because we do that with a very um very specialized approach and yeah we do that with a fully motivated team so please feel free to shoot some questions and challenge me and challenge us everybody is more than invited to go to cognigy.com start you get the full feature set for 30 days no hook um we just just need your name and your email and you can get started and build your own virtual agent and give it a try thanks all right thanks very much can i say silas or silas for english yeah um since this this field is getting more and more crowded of different approaches and and

products what would you say is that that's pretty impressive to be gartner crew vendor and outperforming all the other ones what would you say is the edge of your company or your product um well main reason for customer buying us is we also offer on-prem so we offer sas dedicated on-prem and the majority of our clients really wants on-prem especially in terms of gdpr that is also something we can fully guarantee especially for insurances and especially german companies and yeah the visual editor is like the key feature we have because it really enables business users and decreases dependency from developers because that's a very scarce resource everybody wants to have but not everybody gets understood and since you spoke about your outperforming nlu you have to train that obviously if you do on-prem and if you very highly comply with gdpr how would you train cross-train those models or you simply don't um well we have the nlu how it ships it always has a high performance and then it learns by itself and it's

fully visualized so um that is something you can export and then adjust um but we have an intent trainer where the business users can select okay that's an intent which hasn't been understood and then you can map it to the intent and that's basically how it learns every day incrementally step by step got it and there's one question i forgot to ask a previous speaker eric but there are probably some things that such an ai can never answer what would be your feeling what are the cases where you say no no we step away from that or the engine just doesn't do it yeah i mean let's reference knight rider the ai was able to understand irony and these kind of things and that is nothing an ai can do so that's also my point of relieving agents uh whenever it's a complex case uh too much words uh irony and these kind of things require it we always have escalation paths in our in our tool you should always talk to human if it's

repetitive stupid administrative then a bot can do the work how many languages do you do currently it's 22 okay so covering all europe i guess yeah we do cover europe except for some let's say niche languages for example like slovenia just two million people but if there's a slovenian client we can integrate it quite quickly and since eric made the statement that next year will be the year of the bots what would you say is happening next year from your perspective from the community perspective what are the innovations we can expect um that's a good thing that's a good question what do you have what do you have in the pa in the pipeline for your product maybe um so voice is becoming bigger and bigger um that's very tacky features like specialized features part in and these kind of things and you will voices because we're agnostic so we integrate with yandex nuance uh azure and amazon for example so you can pick any new voice and for example um we have a prospect

with swiss german very special and that will develop so newer voices will develop and then we open to integrate it all right cool thanks a lot silas thanks for your time uh that was really interesting at the end and if you want to reach out to celeste please find his contacts in the app or in the chat great thank you good night good night okay everyone this is the end of the hybrid area it's not the end of rise of ai now we have on the main stage uh more about robots if you like that topic and there will be obviously the final speech it was a great pleasure to host you here with all the guests i really enjoyed it i enjoyed yesterday and today i learned a lot of things thanks so much for listening and good night

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.