I hope everybody had a chance to get a bike at you and a bit of brand food and you and you made sometimes you have a human connection with each other welcome to our AI demo so it's every startup that is chosen we'll have about 10 minutes to do their presentation the first question will go out to the BSA panel and then we'll have one question that goes out to the audience with Abe I think that the topic of focus it makes sense to showcase these companies that are not only writing the trend but they'll also be disrupting it I had the pleasure of supporting Fabien in screening the startup and now happy to announce our final six presentations Congrats to the teams and thanks for joining us of the content on the singularity mindset and a deep adventure into the future so without further ado our first startup besides having a phenomenal football team Barth politics for me feels like a Spanish Berlin because of their diverse culture transformational fast-growing technology sector and lively
innovative startup community similar to Berlin International ecosystem you can actually lose the defeating without speaking well for them there are two local languages I want to introduce our first presentation and kickoff for AI demo we have Catalunya based entrepreneurs justice to explain how they're redefining human interaction through intelligent vision technology so I guess you will remember these film Tom Cruise using Minority Report it's very well known Steven Spielberg's vision was built by my former boss John under Koffler I was the former director of ahh blowing the company that built a software for Minority Report and Steven is silver gather a bunch of scientists and they agreed that in the future machines will be able to see and understand human behavior but this is something incredibly difficult human behavior is complex it has so many different layers it has senses and touches so we decided to focus on one specific problem to try to understand those micro gestures that give you you know the opportunity to interact with the real world my name is Herman Leone
and the CEO justice and we the company that build that software that allows people and machines to understand human behavior and let them interact as you can see in this video these are some of the presentation as in actual applications in the marketplace that very futuristic and interesting companies in digital signage at the multi pen consumer electronics are using today but it wasn't all easy we faced a lot of issues and problems so the first issue that we had with companies we're not ready to understand or use artificial intelligence in this space if you wanted to understand the problem a little bit better current approaches of using deep learning let's say I skip one slide because time's running by you're a company you want to create a new audio experience you have a current approach where you need tons and tons of data to be able to do that and there's the problem which is you need fast product iteration companies don't have the data today to be changing things on an ongoing basis and
the problem gets worse the current approach of the other kind of deep learning means that you need to get that behavior find ten thousands and thousands of datasets model the data and let's get to that level of accuracy so what happens is that if I wanted to create that audio experience this approach would not let me do that fast product iteration and then it will be too long time for me to change the product and put it into a marketplace so it was financially and economically unviable so what did we have to do we had to create a unique method with only five users ten samples each to be able to build a base model that we can put into camera sensors screens and it has to have 90% of robustness so that customers out there will be able to use it and then we deploy life all these installations and we gathered that life data and now we're able to understand and see what users are actually doing so the science team can change
the products and in that same way with the semi-supervised method they can influence the design so if I wanted this gesture to increase volume and this gesture to lower volume but users don't get really get that and they're doing this and that I can quickly change that product life in the marketplace without having to go and collect new data and do many many other things and all of that gives us updated models which allows us to use it for multiple verticals and multiple segments meaning that very rich data set which is life from the different bullet points of the different segments are building better robustness this is our time to data we started with the hardest and most difficult use case which is digital signage tons of people with multiple hands interacting at the same time in malls in stores in buses in any place in any location and we're able to see multiple hands interacting at the same time today if you go to Barcelona and if you're a fan of the football club
Barcelona where for Barcelona clearly you could see a very large video water can take up to 40 people interacting at the same time it's recognizing you it's recognizing your hands and it's allowing you to interact with that what is the base of what I just said what is the breakthrough right here we start with a base model only five users then samples this in artificial intelligence is very novel we can deploy the model locally into the device the device learns itself with live data and then drift tributed this model to multiple products around the world different verticals and that robustness goes across different sectors I'm sorry as you can see I'm actually controlling this presentation it and the the key thing here from a business point of view it's time to application is short and it's economically viable so we can change in a manner a matter of minutes and interaction we can create something magical for users to see and understand that product and this is why we're having success in the marketplace the
opportunity goes beyond digital signage we're just scratching the surface we just signed a fantastic deal with a very large brand that I cannot mention unfortunately and some of the brands that you see here I'm not supposed to tell you about it but this is a small gathering and it's not going to get out of here we've learned that this is a great opportunity the market is ready for this type of technology everybody wants to understand how humans behave in a non mais manner and that's the key aspect of our technology we see you but we don't know who you are and we don't care because at the end of the day what we want to identify all these different humans interacting with technology all you'll be bought into that concept and they're reselling justice into the marketplace for for other brands cisco has chosen to be their solutions partner for entering a lot of different proposals and businesses around worldwide and we are right now dealing a very large installation with 1500 licenses that means
1500 stores that is going to go to 16,000 by the end of 2018 allowing us to reach millions of data point you we closed last year in 520 K this year we're roughly going to get into 1 million point three in bookings but this is just the beginning as I said we have one of the best themes for delivering we have a great combination of business design and Technology we're located both in Barcelona and San Francisco and we come from various areas that complete our team very well to execute on this strategy so if you're interested in experiencing the power of justice come and try our demo and see what I'm talking about thank you Thank You Simon and I'm going to put the first question out for our VP panel do we have any of these things that have a question about the technology or the business concept behind Justin yes George ah yeah go ahead and the question we are Hardware agnostic we worked in different hardware platforms Windows Linux embedded our 4
code is C++ and we're working directly with the camera manufacturers particularly in the space of deaf sensors and in dual camera camera the required depth for certain use cases for instance in car you need infrared because at night you need to be able to see the people so our software uses orbital data that is embeddable and does the key aspect for for firmware in general for the cameras flyers yes I will adjust so as the audience understand the questions and so he asked and to explain how the licensing agreements the worth anything justice we have two models the model for digital signage at the moment works in based on volumes and number of interactions per screen so if you have a screen in a shop that has five thousand people going by and you want to have a short drag-and-drop interaction from a screening to your mobile phone that's a roughly that's a 30 euro a month per point and when we talk about large deals like in the automotive we're talking about a unit
per camera and these goes on and on the millions batches if it's a royalty model here we have an audio question probably just for the future questions maybe a microphone would be helpful so that we can give it out to the audience okay let's just say you're just due to either curious so you guys starting to Barcelona would you say that this is already very receptive there I mean I like the city itself I think it's amazing it's by the beach but also I see that you guys have the diagonal area and a lot of Technology growing there so fast and honestly you guys have so much food and fashion and a lot of business models that would really suit this kind of appetizing platform are you looking to kind of just grow up maybe in Spain and in Europe and then go home or just launch it kind of more global know we we identified that our markets operate quite differently so we're starting in Europe you're using Barcelona as our headquarters for research
and R&D particularly in Europe retail is a huge there's a great market and then we feel that we have we can actually develop it here but for consumer electronics we open an office in San Francisco and we have a completely different operation even a different licensing model there because the markets are very very specific so we're not trying to reach everybody because it's impossible as a platform you need to pick a vertical and go straight into it and our strategy is based on the data that we can get so for now large sets of data are in stores small areas that has a lot of retail and that's why you're across or focus great thank you for joining us today in Berlin thank you so we opened up the demo with a hardware startup and now the next four startups are going to be from Berlin so local startups and software German technology likes innovation the way in like this card fast and efficient for more hip and urban streets hip and urban streets south
streets of Berlin and this next startup will give it will drive us on to the Autobahn of the future with Chris Christmas an intelligent co-driver that speaks languages of English German messaging music navigation and more with the buzz of autonomous automotive trends on the rise German auto labs is doing their part to create a safer and smarter mobility experience for everyone welcome to the stage Patrick good afternoon do we have the presentation all right good afternoon we are German Auto laps we are building Chris an intelligent assistant for the car which helps you to live a connected life while you're in the car in every car so let's have a quick look at the problem we are solving and we will need audio for this otherwise I need to sing which we want to avoid all right good all right so let's look in text form actually what are we doing with the presentation you issue with videos and max utilize them here and snatch oh okay yeah yeah alright three minutes left okay so
let's have a quick look the problem we are trying to solve and now the clicker is not working anymore oh well but it should work with a botnet leaf the decision is new clicker so stick it up a Nevada you can see not often enough left 15 on you mañana - yes man you want to pull someone forward and we fix the laptop in the meantime I don't know you want to bring your left you so while we're getting set up we're going to switch out a couple start us the next one we have presenting is in spurious our modern technology has created so much data but not enough time can we extend time Peter Thiel's seems to think so at least in terms of human longevity but Experian is trying to solve the problem that we can better benefit from what time we do have and putting it to good use this startup will do exactly that it translates the key insights derived from your company's data so that work time is 90% more
effective this means that companies can focus less on data and more and invest more energy into delivering results hmm so thank you very much for the kind introduction my name is Georg Wittenburg and I'd like to talk today very very briefly about discovery so discovery is the topics enticing because there's so many happy accidents that happens when you just go out there like no you go out there and try to discover India but you discover America or a hat ourselves on your head and you discover physics or you leave your petri dish somewhere in the lab and you discover antibiotics and that's that's so intriguing that we add in spirit we are asking the question so how can we replicate this discovery in the digital world how can we leave a like digital petri dish somewhere and then just something miraculous happens how do we get into the unknown and no it's not that you built your discoverer you ships out of blocks but it's rather a question of thinking about hey there's a whole
digital landscape out there that's kind of waiting for you to find the most interesting things on it and this is built out of data of course all right so the question is in this digital landscape we are exactly the points that you should care about we are the decides the interesting bits all right so we are saying it in spiri --nt discovery is really the discovery of new insights in the landscape of your data and it's only half the truth because as you would know you know everyone who's heard Big Data knows that this landscape is so huge that you can't possibly navigate it as a person so here's the twist this is the automated discovery of new insights in the landscape of your data and this is what we're doing it in spirit at infinity building an artificial intelligence that can ingest business data sets tables to think that you guys got in your inboxes as your excel sheets and will automatically scan that data for things that could be interesting for you so
you can actually think about results rather than busy yourselves with doing the analysis and the pivots and all the things that just waste your time and since then I'm actually going to show this to you and so um I was thinking hard and long about what could be a very valuable example and this was the most valuable example that I could find this is the data set from the u.s. American and romantic Environmental Protection Agency in which they tested a couple of vehicles for emissions two years ago three years ago right and so a well-known German automaker is right now putting some eighteen billion dollars aside for saying okay let's deal with the fact that we didn't know that we were kind of the outlier in this so what you can do within speery ins is that you basically you can take that data set you can upload it into our system say start the analysis of this and then depending on the size of the data like the ia will go to work after
some and I pre-loaded this forty five minutes you get something like this so this is selection of presentation slides that the AI generated directly based on the data and I'm going to go directly to slide number four of these results which VI I did automatically for me and this is co2 emissions over NO x and the point up here is Volkswagen right and now you don't have to be a data scientist and you don't have to be a casus to know that you don't want to be that point right and what we do is part of our AI is that we recommend other insights that might be valuable to you so I go to the recommended insights down here at bottom right and you say hey you know these are the top ten outliers on the previous slides like now if you have in correlation you might be interested like know what are the other outlines say hey top 10 folks like and Volkswagen folks are at that point in time now if you're Martin
Winterkorn you'd do something all right and this is the value that we provide and we do this in a very very generic way so just to do this very quickly this here this year this is the data set of the the Fukushima nuclear power plant of the sensor data right and so we can put this data set into our system and without doing anything else it will generate the analysis making sense of the data and saying well you know that's the point in time when your power point was blowing up right or to make it a bit more apply to something that companies would want to do every day this here's the data set from DB Schenker when their shipping containers around the world we can put that into our system not ask any questions but say hey let's make sense of that data here the slides that the system produces automatically and for instance what you get is or you don't see the continents right now but this is basically America on one side
and this is Eurasia on the other side this is the trip that your container takes through the world and you kind of see where the temperature has been rising study of that all right so this is something as a logistics company that you care about because you don't want to have you good to get hard right and so you got a topic of automation in here stuff is happening automatically to your data and you you get to you get to think about results rather than to think about analytics so you know you got the obvious stuff of time and cost saving in here right but what I find most most interesting in the context of artificial intelligence is that AI thinks differently than you right so let's let's reflect off upon this as a human you got a certain knowledge about your business and you got a certain intuition of what might be relevant for your business but AI thinks definitely AI can scale to great data sets so for everything else that is in
your data all right I will bring that to your attention and you can think about that and then you know if we had been around maybe four years ago and say hey Martin Winterkorn there's something in that data maybe you want to have an intuition an idea about doing something you know maybe say fewer 18 billion dollars so um great thing about it fear in this you can actually try it yes someone someone from my team over here so if you want to have a test drive office yourself right here talk to these guys over here and you can just actually roll around in someone else's data if you'd like to and other than that thank you so very much for for your attention every time that you crashed into my house does anyone from the audience have a quick question we only human only has time to take one today because he has another meeting after yes here you have the microphone so go ahead yeah it's a competitive advantage of the mic can
you talk about how clean the data has to be in order for you to produce these life yeah so that's a great question right because in the general like realm of business analytics you have this problem of data a cleansing and you need clean data to run your algorithms that is because you have different tools in the production chain of producing slides that you can put on the table of someone but what we do is we just take imperfect data say hey you know data is missing what not and you just add a sticker to the slide and the footnote saying like we did this analysis anyway on what data was available to us and you actually leave it to the business guy in front of it in front of the screen or in the meeting to make up his own mind to say well is this good enough for my purposes or would I need something better so actually kind of sidestep the problem data cleansing by just saying if you care about this
maybe clean up your data afterwards so okay yep so thank you very much John let's give German auto labs another try let's see if we can get the phenomenon all right the bad news is you will try this until it works okay yes so as a reminder we are building an intelligent assistant for the car I'm practical I said I'm a co-founder and the CPOs or the chief product officer of German Auto labs so let's have a quick problem look at the problem we are trying to solve and hopefully it works this time and it doesn't yeah I don't have one I'm sure they do you want to go ahead yeah you want to okay you know acting it maybe maybe we just look at it this way Kotani for Laura itta for that color gives enough for yellow other aspects and yeah okay let's watch the video hi I'm Chris relax let me send this email and find a better way to the office so if you think this video is a joke unfortunately
it's not and like this 70% of drivers use their smartphone in the car 25% of them actually write emails 44% read text messages and in decades for the first time the number of people dying in in the car per miles driven is actually increasing in the u.s. this has been dubbed the snapchat effect so here's our way how we want to approach solving it and we'll have to pray again but it works hi Chris hi Brian it's Monday are you going to the office yes great this will take 24 minutes please turn on the radio play kiss FM you have a new whatsapp message from Sarah raka would you like to hear it yes please the message is read for lunch today do you want to reply yes then cool half past 12:00 it's all I got cool half-ass pelvic solo should I yes please done Chris running late it's called Peter you have three contacts with the name Peter which Peter would you like to call Peter Abraham Peter male Peter Wolf okay calling
Peter would there's a traffic unit causing a 10-minute delay you can save five minutes with an alternative route okay please take the alternative route okay please take the next left turn Chris please play to Spotify captain you need to talk about you're going to claim all right so there's three ways how artificial intelligence can really help us solve this problem the first and most obvious one is voice touch in the car is not a good user interface no matter how big and how fancy your touch panel is so sorry Tesla drivers it's simply not the right way to deal in the car you will need to look at the screen you need to find the button you need to press it you need feedback if you actually press the right button which in the car often goes wrong that alone takes three seconds easily better normally five and five seconds of the road with your eyes a lot of things can happen if you think voice for the car is a solved problem it's also
not true we recently ran our Kickstarter campaign one of our backers called me up he was very keen whether I couldn't sort of whether this system works for his car I asked him for the specs of his car it turned out it was a 2017 this year bought car a new model with the best infotainment system that this OEM had it had a voice assistant and he said he stopped using it it's simply not working and we're hearing that story over and over again so voice for automotive is still a big growth area and actually or similar to how Jesus earlier presented it's very important to get into that market and be able to iterate fast and collect data and that's what Chris will help us to do the second big field where we need to be smart in the car is actually adapting the conversation to the context you're in while driving it's very different if you're on an empty street and he rode at night or stuck in a traffic jam during the
day and you actually have the mental bandwidth to deal with emails being read to you dictating things consuming a facebook stream consuming any other media you might want to do if you are on a motorway at high speed with dense traffic it's a very different situation and you shouldn't actually be interacting with the device at all and so we're working on technologies which will identify the driving context from sensors from other data from real-time data and actually adapt the interaction to that and the last but long I'm probably even most important one is reducing actively reducing the cognitive load of the driver by making the services easier to consume so here the example is you receive this invitation like I saw in the in the video for going for lunch this is actually not one question this is ten questions because you need to know do I have time where do I want to go how do I get there how do I get back how long will it take how is the traffic will
all be back in time right all of those questions even if they are void space you know it just doesn't make sense if you would be busy three to five minutes in the car looking up your calendar working out stuff so you need actually a system that actually works out all these things for you accesses your calendar and proposes to your options so it says hey I looked up and you're free and you know do you want to meet that you know so ho ho it's like in the video or hey you're actually not free should I propose to go tomorrow or you're not free and you know should I move the meeting around right so these are the three ways how we go about automotive intelligence and artificial intelligence in the car we recently closed our Kickstarter campaign ten days ago we're planning to ship this device by the end of the year to the Kickstarter backers and then it should be coming in q1 next year to an online store near you so
thank you very much please get in touch you know grab me here or give us a call or an email we are very happy to engage thank do you have any questions for the audience and from the audience yes we have a microphone on the floor so just raise your hand oh sorry I'd be interested in the competitive edge as well so why would I need an extra hardware device and why wouldn't I use existing services in place such as Siri or other Google products to control or what you've shown so far yeah so we get this question a lot the reason is they simply don't do the job today it's impossible to do what you saw in the car with Siri there's a couple of issues for example voice in itself is not yet ready which is why we are combining it with gesture with display you need to create a holistic experience for voice to be ready to be a voice only interface that is still a couple of years to come in
general gesture helps you a lot to sort of do things in the car you would otherwise wouldn't do for example disambiguating context in terms of the systems in the car our advantages we can iterate much faster right this is a you know a system that we can update if we want on a daily basis and and can make them better all the time and of course compared to Google assistant or a Siri the other thing we do is we really focus on this automotive context we focus on you know what we call the feat of safety we focus on the context there and we want to provide a better solution in those spaces then you know what you can get from your smart phone okay one more question are thank you for technology so anyways so hopefully over you in the persona creation I mean the technology how the persona works in the high AIS and did you make it yourself or by some technology from somewhere and how long did it take to develop
so it was a bit hard to understand but I think your question was how far we are in the persona creation right so we started actually designing Chris last year in the summer starting actually from persona type questions from interaction models and we've worked our way into a prototyping stage now and you know as you saw we want to be in the market by the end of this year okay great Thank You Patrick thank you we're going to now introduce our next startup paper trail cozy so with paper trails pens pencils and pens for society of millennial minded digital natives this simply does not make sense when we refer to old economy we're not necessarily defining the age of experience but we're pointing to the fact that so for me oh we're doing 20 yeah that the next one was a different startup but we can also start here okay we're sorry there's going to change in order so it's okay flat um we have next up on stage on 20 billion neurons and they
will be talking about the sentient ability of humans and innovating the deep learning potential of machine cool thank you very much okay this works perfect so yes so my name is Kristen - well I'm one of the founders of 20 billion neurons and actually we are computer vision startups so we are still quite new we exist since last year so about for one year and I want you you know here briefly really talk about the text that we have developed over the last year so when you look at computer vision I mean computer vision there are a lot of you know it's important field for last 60 years and now object recognition recognizing CAD or doc and an image is mostly solved so thanks to deep learning which is cool so but this really doesn't happen have much to do with how humans actually see the well so when you have a look at some of these examples which are based on object recognition for example here see an image and a computer produced attack
like horses standing in the middle of the road right obviously there's no horse so this really doesn't work so basically machines really cannot see the world this video and that's really huge problem robotics and a lot of other fields too and a lot of people are working on that so we so that's our mission so the goal of the of the startup that we have is to teach machines to perceive the world like humans so so how come that object recognition recognizing cats and dogs image is solved but these more complex things this more humanized vision is not yet so so there are a couple of reasons for that so basically we found over the last years that the combination of faster computers and GPUs and keep neural networks is actually a great and powerful tool in solving computer vision but you always also need data for that right you usually need large label data and that actually turns out to be a problem when you want to approach my human-like region because you just
don't have to label data so in the case of cats and dogs very easy to label an image and say okay that the cat in the image or darken the image in the case of my humanized version this is much more complicated so basically how do you get data for solving computer vision so you can do other things apart from object recognition so what we have done we have developed a crowd acting platform this is the first of this kind and to give you an idea how that looks like the curse is the first of this kind so I've never seen that before I guess so basically what we do we create certain level labels of complex tasks and X chains and also human emotions or whatever and we distribute these labels through crowd workers all around the world and then we have them film what the labels helps them so for example here we have a label calling which is all falling like a rock and then you can see the video for that
and this is actually great material for teaching a machine about how that really looks like you can't really find this kind of data anywhere else so here's some other example holding notebook view behind a bad frame this sounds really simple but it contains so much information about the world too so basically that you can put object behind other objects that they can include each other and they're still not gone so all these things that we take for granted so this common-sense knowledge about the world this is something that machines don't know and we want to teach it to them but obviously you can also go to human oxygen human action action recognition and emotion recognition so we have a large data set here where you can see for example person smiling or person getting angry or a person drinking a cup of coffee and we have hundreds of thousands of these videos to actually solve important topics and parts of computer vision with that so so for example we use that actually so you saw
a couple of interesting videos there right but actually you can really solve important real world problems with that so one of the first tough that we wanted to target is hand gesture recognition and there we now crowdsource I think largest data set in the world for recognizing human hand gestures and I want to show you a quick demo of this so this is far from perfect I mean it took us about 8 weeks to gather the data train the models and then build the demo so their words we leave something that run really quick and so I can now sums up swipe to left swipe to right I can zoom in with two fingers in zoom out with two fingers and like drumming fingers and which is a bit similar to that shaking hands so I think you get the point what is really interesting about this demo is that there's also the first kind of this kind of demons because usually in hand gesture recognition use the time-of-flight camera this is the depth sensor
so and we are using a standard webcam at a laptop so it also works with a smartphone camera so this is really a step change and the kind of technology that you can use it is really brute force approach in solving computer vision so does take lots of data deep neural networks fast computers and you can solve the problems with that and laughter that that's great so yesterday the SS at so also thanks a lot to our team so we basically build up the company over the last year we're now a team off in total 18 but six you know these are researchers and engineers we are based in Toronto in Berlin so we have a great team icing here two of them also here today Valentin and Aaron so you can raise your hand so if you've any questions about what we are doing please feel free to ask them or me of course and a lot more to come obviously I mean now we also look into the business side of things of
course I mean that's another important aspect of this so we are not a research company but yeah so with that I would like to close our presentation st. you do the desired VP panel have a question first No and the audience um how do you get your samples so what is the incentive for people uploading with you to your platform part two you know what is the incentive for for people helping you by uploading videos to your platform they are they are getting paid for this so so we pay about 10 to 12 cents per video right now and we actually optimize the whole interface around that the whole platform so that is fairly easy for people to generate this video data sorry saying yeah it's the classical crowd Croker platforms okay thank you for more question here where do you get the labeling words so the idea to what to teach to AI all those letters about you say of shaking hands and such yes what do you get to with the source for
this do you have people typing it in or is it art yeah yeah basically I mean this hand gesture so basically we thought about them our selves so we came up with a total of 25 hand gestures and then we made the first video ourselves and this was the instruction video for our crowd workers and then based on that we have generated about 110,000 these videos and the jesters their bid randomly picked so it's very easy for us to integrate new gestures it takes about maybe ten minutes to integrate a new gesture and then you need one or two days to harvest enough data I guess but the question is you have a lot of video citation and I to see what it is yes but you need to write these things so is it done automatically or instead of human writing all those are men I mean the qua quality route or the crowd source video to get you mean the quality of Shuren no no not the quality issues but just what to
do on the video I mean you ask those to do something in the vide are ya getting one so so yes as to so what we do is that we generate this label so it's the other way around right so we generate a label with a specific gesture or specific action for example when we want you saw when we want to target you man action recognition recognizing people are walking or on it we create a label with people working and then people pick up that path they say okay I should make a video where I'm working and then they film themselves doing doing their work that answered your question okay okay thank you so much we're going to now go on tour our next startup is also from Berlin and basically they're solving the problem of paper trails so the world has gone digital but there's still a lot of old economy companies that are having a lot of record-keeping and search Inc is actually trying to make this process more efficient and saving time
and money and one of the cofounders she even started one of her earlier startups in robotics so it let's hear what she has to say about putting very interesting how does it work it fine ok so thanks for having us here my name is Sophie and that's Howard and together we are cofounders of searching so we are in AI phase process optimization what does this mean to very straightforward we go into big corporates but although small companies we look at their administrative processes and we speed them up to 80% why are we doing this what's the status quo so in the 21st century was all about digital document streams fact is that the majority of information that has to be extracted out of document streams is done manually I give you two examples so in order management is over 60 percent claim management over 80 percent expense management over 70 percent so there's a hell of work to do for us a huge potential for cost cutting but of course all the revenue increases so
what is it in concrete that we do we even look at document streams or document pools because we are able to understand what the document is about we are able to bring order in the stream or in the pool we are able to classify and cluster and in the next step we are able to extract the most relevant information for further processing and analytics a visual example of how this looks like this is the clustering process in the next step the extraction of the most relevant information here's an example of our first customer it was a document pool of already 30 million documents here we use already our own developed neural networks because we understood what the document is about the extraction in the next step of the most relevant information and the result is an API export as you can see here so we heard already today how difficult it is to get the access to training data for ground truthing and there was one of our yeah we were very lucky because due
to my previous business we had access to over 900 million documents a lot of them were annotated and labeled already so it was luxury and eldorado for deep learning this robot I invented in 2004 it was quite early for AI it was in hardware but still we are not afraid of types in this space and it gave us a huge advantage because tools like potential flow are open there and processing powers also available we are lucky hooray is supporting us with that and we are yeah thanks if Google is someone who googles here it's a good time to talk about AI this hall is full of people and it's also a good time to talk about clarification we cooperate are not that afraid anymore of thinking about putting the data into the cloud and still 80% of digital document streams and companies are still unstructured so the potential is really really big where are we looking at at the moment ok that's interesting so you have to imagine bigger circles here for some reason you
can see it so in order management today 250 billion euros are spent for exactly this manual transcription process the blue bubble bubbles are the ones where we think that the company is willing to pay us for the cost cutting but now we are focusing on smart process optimization but in the next step of course we will also be able to bring predictions and analytics so the company is able to make better decisions my co-founders team so I'm very very happy to be part of this brilliant team here's the brain behind the neural network we are building together with Eric they are really not only in into research but also into product shipment which is super important it painful discrepancy but due to my previous business that something I'm also very excited about and on the other side we always call them our others we have one guy Martin he's he worked for big corporate which is important for us because we want to enter this world and Stefan is focusing on R&D and financial stuff
we are happy to have people read from Vitruvian on our board and automaker Schmidt he was working for Larry Page for 12 years and is helping us with the hiring process and as you all know it's a quite high challenge to really get the best talent so we are hiring in the areas of HDR computer vision cementing ethology of course all the development we are very very keen on having conversations after our talk we are searching for talent that wants to have a part in shaping this new era and wants to share this big vision with us all other investment side we have really great people I'm one of the startups who is thankful for having and we see brainpower behind us we learn from every conversation and that is something I can really recommend to sit down together with them I listen to them they see so many cases and it's just a big honor to have their input every single time I like this quote would catapult it into the ranks of the
most interesting startups and the world today so hopefully we yea can get there and yeah we are I think both excited of being part of this new era so that's not much time left to have to go very quickly to the technically about the presentation we have kind of modeling the whole process from a document up to the customer database so and start with some classification we can take optically classification so if you have similar documents based on a visual level like pre-printed forms or insurance forms you can also use structured forms that have different layouts like invoices that you have very different layers with always the same kind of structure the exhausting them together be doing then content extractions recognition if there's handwritten text on it if it's printed if it's already clinically we can deal with this and then also very important part is mapping this data to the custom database so for example if you have offer request you have something that's the customer wide on the sheet but it needs to
be translated into the language of the of the company and it's the third big silo of our applications we are using different modules to solve the problems start stop the classification react analysis we are very good in separating static from dynamic content especially if you have reprinted forms very often people write on top of something that's already in the pre printed part but subjecting this part from the document that you get they able do a much better the cognition we have various semantic analysis pathologies behind starts from NLP but also more in the in the instructor document like pattern recognition we are using OCR and HDR the big difference from formatted processes that we are already using our semantic knowledge in the HDR OCR process therefore we know what we have to expect and we have a higher confidence course in the recognition and our latest tool that you're currently developing is despite two independent classifier that will allow us that we remove the problem of meet the need of training different models for different
writing styles big problem handwriting recognition especially there are we are making progress and that's the next big task and the last slide says no ten seconds left there are two main use cases that we call one is the the river the other one is the pond so the one is that documents are constantly flowing in so it's a one-shot recognition you need to extract information and then decide immediately if you want to go with Chado process or manual process and data big use cases it bonds that that you have a big amount of data that can continuously improve and to Quality Assurance with every user interaction you can improve on the result and this requires very different approaches so that and I just had to ask really quick since so many companies are talking about how artificial intelligence will make everything more efficient and a lot of people fear that this might take away jobs obviously not too many jobs since you guys are still hiring for your own team so what's your thoughts on
this for the future our world products like yours change jobs or take away jobs from people we are seeing this more as an enhancement not that much as the replacement so that people do the work that they really wants to do not sitting in front of the PC and just typing stuff in but really focusing on the blog officials too I think we are in the next Industrial Revolution there is no way to get out and it gives us a chance to focus probably also on more human things and our humanity and I think development is part of the evolution we were reducing the working hours in centuries and I think that's more AdWords than a bad thing does the audience have a question and George hello I didn't quite catch whether you already have customers for that but I would be interested in how much onboarding customization and training work you have when you want to onboard a new customer yeah so that's a question a very good question so first yes we have
customers we've made already revenue last year around six seven hundred thousand euro it was project-based so not replicable not scalable of course we know that this is the only way and after my robot company I would never touch again something that is not scalable so now we are in the phase where we just pick projects where we know that the market behind is huge enough for example order management that it makes sense to adapt to horizontal approach we have two one vertically it takes us around five weeks if it take if it takes longer to adapt the technology to one work we would not do it we will focus in summer on the verge cities we are now discussing and yeah for assets of course about scalability the important thing here is that we strongly believe in the main knowledge so we don't think that it's a technology that has no domain knowledge Kendra that's why it requires at some time tools to get into new verticals because we need to understand this vertically but
then just adding new customers it's a very straightforward process and doesn't require a lot of customization okay we'll take one last question from the audience and talking about process automation have you tried reading not document but emails if if I have a lot of emails that has unstructured so the text mining but can you turn that into concrete actions that can be either read by a human or directly executed through the computer we are not focusing on three text so we are using email but only in in processes that already have kind of formalized structure so like an inviting or an authoring process that you have emails with a - PDF attached documents and the Thompson people said the currently not focusing on really unstructured text like emails and really trying to fold energy and it's just a part of our solution okay great thank you guys so much for the presentation and come speak with a message okay so our last AI demo today is our second startup from Spain they come from Alicante
and they are they that they are having robots basically not only robots are learning from humans but for the our next generations they believe humans can also learn together with robots meet a creative machine inspired to improve the quality of life through playing and interacting with our children our startup is a soy and he brought a demo with him is it setup yes okay maybe maybe there there okay a shower under this one you hello hello everyone we are I show you robotics we create emotional robots for for kids which are changing the way they are playing learning and improving the quality of life today we have seen that most many people are worried about artificial intelligence and what we happening in the future with with it and many of them are searching for solution for our needs we are worried about education and how to prepare your children for this future and we think that the best way to prepare them is to improve the capabilities and develop their skills and abilities using emotions
this is the reason because we have created this robot is called I say one kid that is the first emotional Probot for education in the world it means that it's emotional and it it's available to create an effective link between the robot and and the user that it's better for retina better engaging experience it means that is a small robot that had the possibility to made own decisions and they can learn and adapt to a to the to the kid and it means that it's experience stimulus simulates the creativity through play and it means to be a great experience around the robots and an ecosystem with compatible devices or more contents taken in mind the development of the of the children and for the first time we have included in the robot a special system focusing the develop of the multiple intelligence of that of the child we call this system mentor because it tries to propose the best contents for its cut it KITT well let me introduce you I skate you Jose I'm
so sleepy can you scratch my head well so much people hello everybody I'm a nasal robot your social and educational friend I'm a very smart robot powered by artificial intelligence algorithms my mission is to make you happier and teach children through playing games you should know I'm a pro playing tic-tac-toe and trivia I know so many games we can play too I'm an emotional robot depending on how you read me I could get sad you can use me to learn coding using for example scratch usually I get surprised watching how fast you progress or even happy if you make me feel good for instance playing with me when I am happy I enjoy dancing check my slow amazing don't you think so I can understand you and have a conversation feel when someone touches me and even see you with my camera you could say I'm a gossip but the thing is that I love meeting new people well thanks to my camera I see there is a lot of people in here I'm getting
nervous sorry but I'm quite shy you know I love what you call in earth music I wish that would exist in my planet Kazushi you can send me your preferred music and I will play it for you I love listening music with you if you hold and drive me I can feel that I can fly why did you stop well maybe you don't want to watch me flying as you have already seen I can move using my server motors play music and sounds and show emotions using my belly and mouth I spent a very good time with you I'm sure we will be very very good friends see you soon what this is a su Guang we have developed a platform around this device with several engines the most important is the emotional engine that is the core of the whole platform and we put this this engine in the service of other developers or to you too to build a ecosystem around the robot our community is growing the array we have more than
3000 users mainly in Spain France and us and in the next year we stand to to UK probably I would like to to show of an a video for showing how to interact the robot with wicked scene in a real contest No well I'm the last thing that we like to say is that we we believe that education is not possible without emotion thank you so much