 Thank you and thanks for giving us the opportunity to talk about these use cases on Microsoft Cognitive Services. Just to give a quick introduction about myself, I have been working with Microsoft for about 15 years on different technologies and today I am going to talk about Cognitive Services which is quite interesting and why it is interesting is because of these three things. The first one is about the compute and this is under AI track so I want to talk about that. The compute is available today with different cloud companies coming out including Microsoft, we have Azure and then many many others like Google and Facebook and others were building on top of it so the compute is easy, scalable and we have powerful algorithms which has been developed for quite some time now so including on top of it we have this massive set of data. So what is Cognitive Services? It is essentially built on this three core which is around AI services or APIs which allows you to build applications very very fast using AI and all these interesting part of it. So I am going to talk about different touch vision, speech, languages, knowledge services, speech and search and labs. There are few of these categories I am going to talk about and I have some demos which I will walk through very quickly in this timeframe. So let's start with what exactly what we provide in vision and other scenarios. So while I wanted to come back to the slide, let me start with some demos and then I can come back to the slides and talk about those quick cool things. So let's start with a very quick demo and most of these things are available online on GitHub and I have kept that in the slide so you will be able to get to these code available and play with yourself on this particular Cognitive Services scenario. So let's get started. I am going to get started with a very quick face recognition part which is very very common but at the same time what it brings and entails interestingly is that it is being used by some of the cop companies and some of the startups and most of the middleware companies who are using it internally in their products. So let's get started with the face part very very quickly. It's a very standard system. If you notice there are a few things which will be happening. Actually my webcam is predicting and it's also looking at the projection and it will be finding me out. So I take a quick picture of myself and it will tell me who I am of course and it also predicts age which I would love to be at 33 but I'm not but it also tells me it's a 77% accurate. So it's not fairly 100% true but it's taking care of the false positive and other scenarios. So this is a very basic thing which I think most of the people have out there. Now let's get started with where it could be used. So in retail scenarios for example where we are using it and we are intending to use it is to look at footprint counts. So maybe I request a few of you to come here if you can please. If you notice what is happening here is it is telling me who I am and how many person it has been recognized. Now it is able to recognize the host as well. And if you notice if I smile or if I make a sad face it will go ahead and tell. So there will be a marky at the bottom which will tell you what emotions it is. Thank you. It will go ahead and tell us what kind of emotions. So it is not just about face it's also about emotion and where it has been used. So in Singapore we have been working with sync polls which is to look at people who are real and be able to register them especially for citizens and PRs because gambling is restricted and it's also legal but for a certain point of time. So what we would be needing to do here is to be able to identify those people and there are a few categories through which we are able to identify. So I'm going to stop this application for now and run another one which we have created and come back to the slides a little later. Just wanted to give you a perspective of how we are able to identify different users and some of these services as I said is built on top of computer vision. One was face I showed emotion. Let me talk about and show you a quick demo on the OCR piece of it. So there is this nice small sample code which I'm running around BOT which will be able to OCR your ICs and just to do that very quickly let me just go ahead and connect. So there's this tool called BOT emulator. Alissa who's here did a nice talk last night on BOT framework and I think you know it's recorded so you'll be able to see that. I'll just quickly go ahead and talk about it. I'm not touching based on BOT but I'll really say hi to this BOT just to say you know what it can do. It will tell me that you can attach an image and it will do it will find words in that images. So just to give a quick perspective of it, let me try to do this. I have few images of some of these ICs and if you notice what will happen is it will go ahead and connect to the cloud service. It's basically calling a computer vision OCR service and it does OCR for me so it tells me it is Singapore identity with the identity number and so on like all the details of it. So it's quite easy now you can actually go ahead and take this. You not only are able to verify the person and because we have a face ignition so you will be able to catch who the person is but you also get the data and then you can validate the data. Now on top of this we have another service which can use to customize this to find out if this is really a Singapore IC or not. Is it something belonging to some other country? So what I've done is I went ahead and trained a service into this. It's called custom vision service. Instead of using ICs I use passports just to give a quick demo of this. So this service is also available. It's free for now. It's in preview mode so you can just type in customvision.ai and you'll be able to find this. What I did was I just used some images online and trained some of these guys. So I trained for Malaysia, Indonesia, Malaysia, New Zealand and Singapore passports. The minimum category you would need to, a minimum number of photos for each category would need to have is five so it will be training itself. Notice what it is doing is these images are not really properly scanned. These images are photos. What I did, I already trained it to save time but it takes only a few seconds or a minute maybe depending on the size of the images. So I don't need to train this anymore. It's trained already. So I'll go ahead and take a quick image and show you how it functions. So I'll go ahead and take an image which is for test scenarios. I have kept some of them in my folder. Let me just go ahead and take it out and show you in custom vision, test images and let's take this picture. And this picture is not there in the trained image set. It will go ahead and tell me different predictors. It tells it's about 31% Singapore passport. Notice what is happening here is because the passports are not scanned properly. It is just an image. However, it will still tell you what probability it is. And if I have had more Singapore passports trained, I would have got a better result out of this. So now imagine a framework which could happen. A framework wherein you automate this authentication processes. So you have face recognition. I talked about documents. You can actually validate the document automatically and then scan those documents. So you're not able to only validate by the photo but you are able to look at the content inside those photo and validate that as well. And then of course we can take entries of that and make an entry as well. At the same part of time, what I wanted to show you here is I'll go back to that demo which I did before and talk about this kiosk application which is also very interesting. Let me just close this first. And I'm going to go ahead and take that application back because there are a few more interesting part of it which I wanted to cover before we go back to some slides and talk about those. So talking about vision, another scenario where we have been using this computer vision is at some of the malls and including companies like S&S. So if you notice what is happening here, if I'm in front of this kiosk, it identifies me right away. And to be able to do that, what I have done is I've used this computer vision API and I've just used the keys which is on the cloud. And I'm using those keys and I've trained myself very quickly. So there is a setup which you can do. This application is also open source completely on GitHub. You can go ahead and find it out as intelligent kiosk and you'll be able to look at it. So I've trained myself here. I've added some pictures. In fact, just two pictures out there. But if you can add more pictures, you'll be getting much better result. The advantage of this is that even I don't really look at the camera, facing the camera, I can actually look left, right and it will still tell me who I am and what kind of image I'm looking at, like who I am and am I authenticated or not in this case, right? So a quick example of this where we also used, in fact, my colleagues who worked on this project are also here, Alyssa and James, sitting there. We have this very interesting video analytics part as well. So we can not only analyze the photos and this is a real use case where we used in Sabana Jurong. This is a lift and there are a few of the colleagues from Sabana Jurong and some employees out there. If you notice what is happening here is that, and I'm running this on a phone network, so this is connected with a phone, it will still go ahead and tell me, identify who the people are by processing the frames and then what are their emotions and also predict some image, you know, age and other scenarios. So some of the use cases here, which could be around just using these cognitive services and we could build applications very, very fast, is if a person who is known should be within the compound or not, if a person who's blacklisted should be within a compound or not, like the moment you identify the person who is, and in fact, in this case, we have not trained the pictures of the person, but we are able to train the pictures, like just give photos of people who should not be in the compound. We can easily notify right away the security team that, you know, these are people who should not be there and should be escorted out, right? So some of these are very, very interesting to be able to use and build quickly. Just to give some more perspective on this from the vision perspective itself, I wanted to add a little bit more on this chart scenario, which is also very interesting. So we do a lot of work in office and we use different charts and graphs. So let's look at quick example. I'm taking a quick sample of this, right? I'm putting an image of a chart and what it'll be doing is it'll be analyzing this chart and it can read it out as well, of course. So let me just point it out to machine. A pie chart title most popular dog breeds with six slices. Shepherd at 30%, Beagle at 18%, Labrador at 17%, Retriever at 15%, Bulldog at 14%, Poodle at 7%. So it's quite interesting to look at, right? In terms of charts, you can leverage this as well in your offices. These services are built very quickly again. I mean, few lines of code in fact to write this whole thing. So I quickly wanted to do a walkthrough of the code as well, right? And this is Xamarin application. Xamarin is again open source and available on all platforms. So if you look at it, what I'm trying to do here is basically call few services. Some of them are computer vision service. There is a face service and emotion service. All you need to do is like npm packages in Node. We have Nougats. So you can go ahead and use those Nougat packages out there. Some of them are here very quickly just to give you a perspective here. It's the face, emotion, and vision. And then there is a common API, a common set of API which I've included in this particular code. And then of course you have to use them. This is a bad way of programming. You don't have to put the configuration out here. Just I build this demo very, very quickly. So I put the code here. In fact, there is a good sample which I've posted in my GitHub. You can always find that out. In this case, I'm just looking for a picture, a camera button, and then upload the picture. What we're looking at is the face ID. This particular image will be returning us a face ID which is out here. So let me just open up computer vision face recognition API and show you exactly what it does for us. And I'll be running through some of these other samples as well because we talked about OCR and other stuff. So if you look at this, for example, I took a picture of Satya and if you notice it will tell me. So it returns these values in a JSON file. It also returns a GUID which you can capture and then you can validate. So for each faces or different types of faces, you'll be returning a GUID which you can validate to see if the photo matches or not. So just to give you a little bit more perspective on this, I'm going to run one more sample which is also on GitHub on the link. And I've added all of this here on my slide as well. So when the slide is out, you'll be able to find all those developer links. So what I'm going to do is I'm going to very quickly capture a face folder which has some of these photos and let's look at this what it does. So it will automatically arrange all my images with different types and so on. It will recognize all the faces. So if you have pictures of two people or the same person twice or twice, it will automatically be able to list out. But it not list out only with faces, you can also have emotions. So for example, I want the happiness people who are happy. You will be able to see all those pictures out there. If I want to say people with neutral faces, you'll be able to find that out as well. And of course, I want to look for animals. You can get those pictures out as well. So it does quite nicely recognize from the vision side. And I've been talking about vision for quite some time. Let me just go ahead and talk about a little bit more on the other aspects as well which is speech. I want to very quickly cover and then I also talk about video which is very interesting as well. So on top of what we have from vision and face, I wanted to talk about this new service called speaker recognition which is quite interesting as well. Because the way we talk, everybody has a pattern of speech and you can identify from speech who the person is. Of course, this is not 100% accurate. You have to train it more. But just to play this audio and quickly show you, these are for pre-trained audios. It should be, I think, let me just refresh this page. It should be able to play the audio and tell me who the person is. I think the audio is not coming out, but it's okay. I mean, you can try it out yourself. This is very much available on Microsoft cognitive services and speaker recognition part. Okay. What I want to do is I want to want to talk about this video indexer as well because video indexer is quite an interesting service which you have introduced recently. You can upload videos and then you will be getting all those metadata even without pre-training it. So in this case, I have not trained this video. I have just uploaded this video and I'm going to go ahead and play this video for a few seconds and then I want to show you what exactly we can do with this video. So while this video is being processed and what you can see on the right-hand side is a set of people who are out there who are in this video. So while this video is being played, and I know you can't hear the audio, it's fine. You see that there are a few things which are coming out. I'm going to try and replicate this. All right. I mean, so this video is being played. There was OCR text and then there are some people in this images. And what we are able to do is we are able to identify. I mean, I have not trained with this photo. Again, if you train with the photo, you'll be able to recognize who the person is. But at the same time, there are four people. You see that, you know, we are able to find out the keywords specifically and we can always go back to this location when it was. So what I'm doing is, right? And at the same time, you have annotations and then, of course, we also get the speech sentiments. From the speech, we can also find the sentiment analysis and we can say if it's positive, negative, neutral and so on. Now, I want to show you some more interesting stuff. So look at the transcripts. So we can generate the transcripts out of the box right away without pre-training the videos. You get the transcripts. In fact, the transcripts also have text which are OCR, right? So in the video, we have this separate layer for text and that could be OCR. You know, you will be able to find that out. And not just in English. There are many, many, many languages which are added out here. So I don't know how many of you speak, like, for example, simplified Chinese. Let me just go ahead and show you very quickly that. Right? It will go ahead and convert the whole text in near real time for you, including the OCR part and the text and you will be able to see this whole text quickly. Now, that's the power of cognitive services. Like, we have pre-trained set of APIs which you can directly go ahead and embed into your code by using some of these. If you are using Node, you can have NPM packages. If you are using Xamarin C sharp, you have Nuget packages. We also have these APIs available for other programming languages for Python, for example. People use it quite often. So these APIs are available on cognitive services. If you just search for Microsoft cognitive services or go for Azure and cognitive services, you'll be able to find these. All these are available on the link. And then on the GitHub, we have published it. So the slides will have it as well. So let me just go back and talk about this, wrap it up. I have already shown some of these slides. So I'm going to quickly say why cognitive services is very, very easy. It's flexible to use it in your code and has been tested out. So we have published it on GitHub, Stack Overflow, MSDN, and so on. So you'll find it out very quickly. Then the links are out here. So I've already done some demonstrations. The links are out here. You'll be finding it out in the slides. So you'll be able to take it up and build on top of it easily. So with that, thank you and have a wonderful FOSSAsia event. Thank you.