 Okay, there are already 130,000 users. So if you are starting new project, if you're working on idea, go and check out the community. And there is a Python model, which helps you to talk with things speak. Let me show you how things pick looks like. This is how I'm just locking and using my credentials again. It's free. So you create any new channel. It has eight fields. So you can get data from eight different sensors. If you have thousand different sensors, you'll have many other channels. Okay, so I'm just opening up one of the channels, which I've already created. It has, you know, okay, this is located. We can have locations. It is in the US wind speed, humidity, temperature, very nicely captured continuously data is is getting in. Now, let's see you want to do some analysis. Let's say due point calculation, there is no sensor available. You can do analysis right there. So they are ready made templates available. You can create your own template and you see here, it's not very clear, but I'm looking at a channel ID. It has API key and then writing simple scripts. It's for the generation language, if you have not used it, okay. If you have MATLAB and want to do more, you can import data from things speak to MATLAB. So there's something called things speak support toolbox, which is available in MATLAB community. Okay. So MATLAB and things speak. They are tightly integrated. Now, what you see is a MATLAB script. If people who have not seen MATLAB, this is how MATLAB script looks like. We have turned into live script. Okay. So it has text here. I'm saying I'm reading six days of data around 8000 points and doing basic stats. Now you see here, when we have, you know, data, which is daytime data, it really makes sense to have in really nice fashion. Okay. So there are special data types available for dealing with dates and times. And then, you know, nice equation and why I'm stressing on these points. Documentation also is very important. Okay. And if you write this nicely, you can actually just generate code or generate a HTML page. Okay. All right. Let's see. Have got this code ready to take it to cloud copy and paste control C control V. Excess. Okay. So we started with challenge. We want to develop IOT quickly and easily. This is how you can do it. Okay. So we had data. We did not have a particular sensor. I just calculated that and I have that ready now. I want to check particular data every five minutes. I can design a time control that I want to calculate this activity or check maybe a temperature sensor on your greenhouse every five minutes. You can do that. So I'm here saying that I'm deciding the frequency every five minutes run that particular script, what you have developed. What next? All right. You've got the data. You're continuously monitoring a particular signal. You want to also act on something. Temperature goes beyond in a factory turn on the fan, correct? So that's where the reacting piece comes in. So you can. So for example, here, this example shows that your mobile is continuously giving location where you are as you come close to your house, turn on living room lights, right? So you can react on it as well, collecting, analyzing and acting on the transmission or data, data transmission to think speak is very secure. We follow HTTPS and impurity protocols. There is also got to have a matching API for it. Okay. We talked about the cloud. How do you make or enable your devices to talk to the cloud? Okay. So for that, I'm going to show you a few blocks from Simulink out of Simulink. Okay. Simulink is a block diagram environment where I represent physical blocks in terms of as in physical systems in terms of blocks. Here I'm showing you various different Android sensors. Let me just zoom this out. You see here, so various different sensors, accelerometer, temperature. I don't have to write script for this. Okay. So let's say if I want to send accelerometer data to think speak or cloud, drag and drop blocks available to think speak. And I'm showing here or telling Simulink that I want to run this directly on Android. Within few button clicks I have or I have turned my mobile into a edge node. Makes sense. Okay. This support for so many other hardware devices. I'm connecting back to the challenge we said, you know, you've got to be an embedded engineer. I don't think so, right? If you want to know more how to get your audio connected, there are tutorials available. Okay. Let's talk a bit about developing predictive models. Now Mr. Chabriwala, you know, touched on this particular part just because you have algorithms, don't use it, right? Then and then machines, what algorithms can do better. So when problem is too complex, we need to adapt to changing data and when it to scale only then use machine learning or deep learning because algorithms can do that better, right? And this is a generic machine learning workflow. As you know, my highlight here is that this is not one time approach. It's iterative. And once you're happy with your model, then you go ahead and, you know, use the model and new data. Now if you are new to machine learning, that was our second challenge that don't have to be a data scientist to build predictive models. There's a lot of documentation on concepts as well as if you have 400 plus features and don't know what feature is really the right signature for distinguishes in class, you know, various different activities, then you'd be using few of the inbuilt techniques like NCA, which will help you to automatically reduce or select right features. Okay. Okay. Okay. Machine learning. Two questions I used to have in a few years back, you have so many models just because I know a particular approach. I should not choose it, right? So which model is the right model and how do I make it sure that the chosen model is the really best model for the given data set answer, compare many and then choose one, pick one and then do hyper parameter tuning, optimize it. Okay. Let's see this. So I'm going to open something called classification learner app. Okay. Open this outside. All right. You see here, I'm just opening an app. And. Okay. This is not playing as I expected, but it gives you action. So I'm selecting training data here. I have complete control. What's my predictor? What's my feature? I can select that interactively. I can define what is my validation method. So when I keep 25% for testing, you can do that interactively. Okay. Once that is done. Okay. Let me just go back here. So there is a. Complete drop down where you get to select all the machine learning models. Okay. You see here. So there is a drop down. If you don't know, or if you have completely new to machine learning, no harm in selecting all towards then the machine computer will do it. Okay. And you can do this in parallel. So what we're saying here, our key point is that you don't have to write your own algorithms. And when this is training, this can happen in parallel. Okay. And once you're happy with it, you can go ahead and. Just go back. See, so it will recommend what is really working well. There is a confusion matrix and you can generate code out of it. So again, there is a lot of time saving. Okay. This is not black box. You can very well control that or tune that. All right. Next challenge was how do I make it sure that the model is really the best model? Okay. How do I make that better? I'm doing hyper parameters. Many of you might be knowing these are internal parameters. For example, for SVM, it could be box constraint or for a classifier. It could be learning rate. Now, if you control these, you get more, you know, or better model challenges that this is time consuming. Okay. There's something called Bayesian optimization and you can do that automatically. So. Okay. So there's some problem with the resolution. This is not getting played, but point is that you can actually do this automatically and it will give you the best hyper parameters. Okay. So you're very sure what you have is the right model. Point is that there are so many other apps which will help you to do your data science really fast. That's our objective quick and easy. Okay. And if you are very new, there are a few white papers on choosing the best model and avoiding overfitting. I'll suggest to go ahead and read it. Okay. So we have our analytics ready. Next is either it taking it to cloud or taking it to embedded target. Correct. Makes sense. Now you need to write special code. You might be developing analytics in one program and then you write or generate a code in other other language. Right. How about generating it automatically? And you can use something called matlab coder. There is a facility few button clicks. Take the code available and it will generate give you a C code, pure C code. You can go ahead and change it. Okay. Modify it further. So what we have done taken the machine learning model and generated code for it. And then it was integrated with with iOS on left hand side. What you see is mobile screen and it is detecting the activity. Okay. So he's my colleague set was developed this is sitting and it is automatically getting recognized. So these are very important parameters again. I mean, you know, 20 minutes is very short. But what sensors you will consider for detecting activity. It could be gyroscope. It could be accelerometer, but it should not towards then eat a battery. Okay. We are touching on those pieces in this talk. But those are again vital pieces. Okay. Uh, quick quote. I mean, you know, people ask whether it's really because, you know, towards then automatic is automatic. So here would have phone group R and D says that it was really error free and it saved six months of development time. Okay. So taking analytics, what you've built to edge notes or cloud is really easy. Simple. Okay. Uh, if you want to take it to cloud, we said that it was control C control V in case of things speak. But if you're using other third party, uh, you know, clouds like Amazon or Azure, you can very well do that using Matlab production server. So if you have further questions, I know this is a lot of material in 20 minutes. We have a booth and will be available two days to talk more about it. This is what we started. Naughty is not easy. It requires a lot many skills embedded engineer, compute, you know, cloud computer, as well as data scientist. And we also said that, you know, being startups, being new project, we have limited resources. So you want to achieve these things quickly and easily. And in the short 15 minutes, this is what we saw. Things speak. It's a data aggregation and analytics platform. You don't have to be a cloud architect. No time and money spent in creating web infrastructure. We also looked at automatic code generation. So programming embedded devices. It's very easy. You can use Simulink to represent physical system in terms of blocks. Okay. And for data science, there are apps and documentation, which will help you to do things really fast. Result is prototyping becomes really quick and easy. Now, again, we started also with one of the challenges on startups. So if you are a startup, do and talk to us. There are already 2000 plus startups around, you know, 80 startups from India who are already using MATLAB and Simulink. Sorry, I'm blocking your video, but they're reducing their technical risk and doing more with the limited resources. There's also special, you know, program for startups and accelerators. Now, I always get motivated with startups and let me end this talk with awesome thing built by one of our startups. Autoscope, if people don't know, Autoscope is for ear inspection. Okay. So Cellscope is a company. They had, they have used just a mobile phone and an attachment to build a device which similar like similar to Mr. Chabrivala, you know, what he had done. So let me just open this link here, who it plays outside. Okay. It's pretty slow, but what they've done is this is a device and it happens to me when symptom is there, when I visit doctor, symptom is no more there. So capturing symptom, when the pain, you know, it becomes really easy for doctor to diagnose it faster. So this is what they have done. Image processing is happening directly on the phone. Again, we touched on that particular piece, you know, in the beginning that what runs on the age node and what really runs on the cloud. If you have, let's say if you're building smart city project and maybe a nice signal management system, and there are a thousand different age nodes with camera, video feed, you won't be sending video feed to the cloud. It will eat up all your bandwidth. Right. So this is something is done. This is pretty awesome, which I really get inspired. And I'm, you know, yes, yes. So this is my last slide. I wish to know what's your awesome thing. Okay. Thank you very much.