 I'm Usha and I'm a Principal Data Scientist at the AI Research firm, my pseudo consulting firm, and I'm also the Bangalore chapter lead for Women in Machine Learning and Data Science, and I'm also the Ambassador at AI in Mac. So the topic we're going to talk about today is deep learning to fight hunger and poverty. There is three speakers lined up for this talk. I'll start with a quote. This quote has been told by Theod Skulls in his Nobel Prize Economics lecture, and most of the people in the world are poor. So if we know the economics of being poor, we would know much of the economics that really matters. So this was told in 1979 and now we are in 2019. There's several years which has part, the statistics has changed and things have changed considerably. So in 2019, if you see the current statistics, we are in a much better situation compared to what 30 years back, the situation was for relatively poor people. We are at a much better situation, but still if you consider the relative gain of the people who are extremely poor, that is not progress well. But the situation is much better, and introducing the most important concept, poverty trap. So if you see the poverty trap is a very vicious cycle. So if you see violence and low productivity will cause poverty, and when there is poverty, there's a lot of hunger and nobody does agriculture or nothing. So obviously there's a degradation in the natural resources, and obviously hunger will initiate violence or instigate violence. So this is a vicious trap. So it just gets trapped and it's very difficult to come out of this trap. So that's a talk about how can we use innovation and AI and ML techniques to cause disruption to get out of this vicious cycle. So the first part, the first 15 minutes of the talk will focus on how you can use deep learning techniques to first identify the poverty zones. So it's very important that you identify the zones, only then you know exactly how to focus your resources and volunteers and subsidiaries to that part. So identifying these zones is very difficult and there are a lot of challenges. The first 15 minutes of the talk will focus on that part. And the second part is once you've identified these zones, how can you use innovations in these areas to combat hunger and poverty? So the third part of this talk is you get a lot of insights from AI, ML techniques, and what can government do with these insights? How can you come up with better economic and technology policies, which will combat, you know, how can you bring about better food security laws with all these insights? And how can you help ensure that such laws, you know, enables wider adoption of AI techniques? So it's not just in one street you adopt these things. So government has to come up with policies such that it enables in the wide scale adoption. So that's the third part of the talk. So we are three people from diverse set of backgrounds. So yeah, so my talk starts now. So deep learning techniques, why, how has been, how are we taking statistics prior to deep learning and ML? We have been witnessing ML revolution only for the last 10 years or so, maybe before in a different name. How has statistics been taken before all of it? It was basically through door to door survey and, you know, getting a lot of volunteers to go every door and take all these survey. So this is a very, very expensive operation. And so a lot of funds is being allotted in doing all these traditional surveys, which in turn can be used for a different, you know, much more better purpose. So it's very expensive and the data which is being collected is like, has very poor resolution and not, is not very useful. And there's a lot of, you know, lack of ground truth as well. So, you know, the traditional modes of survey has a lot of disadvantages. So this can be better with deep learning techniques. So if you see satellite imagery is very cost effective and it is easily available in NOAA, you get satellite imagery every 30 second. And these are easily available at very cheap and you can download it. And they're much more accurate as well. The insights which comes out of these techniques are much more accurate than your traditional surveys, which is very labor intensive. And using satellite imagery is a lot of techniques which has come out. But there are like the approach which I'm going to talk about is a transfer learning approach. So you have a lot of nighttime satellite imageries and daytime satellite imageries. So the nighttime satellite imageries, what are the main challenges there in the daytime satellite imagery? It's very difficult to get a label data. So, you know, given an image, it's where you need a lot of people to sit and label which is road, which is lake. So it's very difficult. So number of labeling, label points is very limited. So the scarcity of data, that's where the transfer learning approach helps. So this is one of the breakthrough paper in 2016 from Stanford University. It's from transfer learning from deep features for remote sensing and poverty mapping, but Neil and Jane and Steve. So this paper, this baseline model is used by a lot of people across the world, across countries. And there is a bit of optimization done about this model. It's not they use the exact model. They do a little bit of optimizations over this model. And this is proven to outbeat, outsmart, much other, different other techniques available. So this is one of the best baseline features we currently have. So I'll just quickly walk through the process of how this is being done. So we break down the problem into three parts, problem one, problem two, and problem three. So the problem one first thing is I train, I take my VGG 16 model and I pre-train it on ImageNet. So the first problem is object recognition. And the second problem is I, from the daytime satellite imageries which I can download it, I can, both nightline and daytime satellite imageries can be downloaded from the corresponding websites. What I'm trying to predict is night time, light intensity from daytime satellite images. And the third problem is predicting poverty from the daytime satellite imagery. So I'm just going to quickly walk through the notebook of how exactly this is being done. So the second problem is actually an intermediary to achieve my goal. So I break it down into three parts. So from NAOA website, I'm going to download the satellite nightlight images. And then DHS is the most trustworthy survey information, which is currently available for Africa. So I download the DHS data for a particular file. Most of these surveys has been done in five of the African countries. And Rwanda is one of them. So I'm going to download the DHS data from Rwanda website. So you just have to go to the website, register yourself, you'll get a key, and then you can download the data from there. And then the third step, what I have to do is I have to test whether the nightlights can predict well. So before getting into the transfer learning approach, first I'm just trying to predict the belt using nightlight satellite imagery alone. And then I'm also trying this approach of predicting using daytime satellite imagery alone. And then I'm using the transfer learning approach too, so that I can compare across these three approaches to prove transfer learning approaches the best one. And then finally, I'm just constructing a map which just shows the belt distribution of Rwanda. So this map I can just take it to the policymaker, like he's a Mackenzie economist and affiliate professor of university. I can take it to people like him and then, you know, this map will give him enough insights to, you know, to work on the policies and economic and technology policies of it. So let's get started. I'll just quickly walk through the notebook and how the process instead of just giving you a theoretical statement, I'll quickly walk through the notebook. So this is, so I'm downloading the first step, downloading the nightlight satellite imagery for the particular state Rwanda. And once I download this nighttime, I'm also downloading demographic and health survey details for this particular state. Once I have this demographic health curriculum, there's a particular column called wealth. So, and there are different clusters of Rwanda, different provinces comes under different provinces. For each of the province, I'm just finding the centroid. And so the first step, as I told you earlier, first I'm trying to predict wealth using the nighttime satellite imagery and this is what has been done in this block of code. So I'm just the DHS level data will give me the belt aggregates and I'm just merging these two nightlight imagery data and the belt aggregate. I'm extracting some important features like the limosity of nighttime light and certain other features like mean, median standard deviation. Once I extract the important features from this imagery, and then I'm just trying to plot a graph between the nighttime limosity and belt. How is it correlated? Trying to fit a regression line between these two things. And I'm doing the same thing for daytime satellite imagery now. So if you see this daytime satellite imagery every, I'm downloading it from Google static API library. And so this has some usage limits and stuff. Maybe once or twice you can download it for free. So I'm downloading it as a 400 into 400 pixel and each of this image corresponds to one pixel in the first image which I showed you, which is the nighttime satellite imagery. So once I download this, I'm doing the same process which I did before, trying to predict the wealth using daytime satellite imagery now. So I'm trying to extract some important features and then the DHS data contains the wealth for every cluster it contains the aggregate average wealth. So I'm trying to like, I'll extract some features from the daytime satellite imagery and then I'll merge it with the DHS data and then I'll plot a graph. So this is for my second model, which is just getting the insights from daytime satellite imagery alone. And then I'm using this transfer learning approach. Transfer learning approach, basically I'm using this BGG16 model which is pre-trained on ImageNet. So one thing which I'm just altering out of this model is I'm converting this fully connected layers into fully convolutional layers. So that is one thing which I am making a change in this model, existing model. And then I'm just doing the same thing which I did for the first two experiments. So this is the, I'm just doing this for that transfer learning approach. And transfer learning approach, I'm getting a better accuracy compared to the first two approaches. And then I'm trying to finally build a wealth map of the entire state of Rwanda which can be used by the technology policies and stuff. So this is just a quick walkthrough of the code. I'll be putting all the codes and the slides in the proposal, under the proposal section so which you can go back later and dig through. So the transfer learning approach has been always found to outsmart the other approaches. And it's as quick as as close as to manually going and doing a survey door to door. Is that accurate? And it's very cheap and very flexible approach. So this ends the first part of how you can use deep learning and ML techniques to identify the poverty zone. So identifying poverty zone is just one part of the first part of the pipeline. So once you identify these poverty zones, what are the AI innovations you can bring up in these areas to combat hunger? So poverty and hunger are the first two sustainable goals development of United Nations. So the next speaker, Shalini will talk about how to use ML techniques to combat hunger and poverty. So to Shalini. Thanks, Susha. Okay, so I'm Shalini. I lead the data science team at Numerify. It's a US based startup working in ITSM analytics domain. So as you can see in this slide, what Gandhi told long ago, this is still holds true, but it's still we have approximately 800 million, 800 million people still staying hungry every day. And with UN, as per UN, the global population is estimated to hit 9.5 billion by 2050. And if we, with the current growth rate of agricultural products, this will change by 2014. So this is where we stand. And we need to basically a profound change is needed in the food and agricultural system. And that's what I'm going to talk about how AI can change, how AI can contribute to this next green revolution that we need to meet the demand of global population. So the first in the list is there are several AI innovation in agricultural field in the sector of agriculture, but this one stands out, which is a project by CMU. This is a precision farming project. They have a comprehensive system of drones, sensors, robots and AI to actually identify for selective plant breeding and better crop practices. So what they have done is they have created an autonomous, basically a robot, which can take a visual survey of the entire crop field and it uses a laser scanner and laser scanner and a multi-spectral camera, which is capable of taking basically invisible radiation spectrum. So it takes that and then once all these, the scanner can capture plant geometry and the multi-spectral camera can capture all the say plant functions, plant parameters under different climate conditions. That is what is used, then the mathematical, basically AI model is applied on top of it to identify the traits of the crop, which is most suited for higher yield and most resistant to drought and heat and diseases. That's what this project is about. It's basically to grow, there are several application, one is to grow a protein-rich crop called sorghum, which can be basically planted in heat and drought-prone areas. And then back in India, here in India, we have a Wadwani group of Wadwani AI. They are experimenting with our best AI-based, best control system, which identifies a pink ballworm, which actually causes, which eats the cotton ball and causes huge damage to the crop. So while there is one angle to create like higher production of agricultural, the other side of it is that crop damage cause severe poverty for the poor. So that's another angle and that's why all these best control systems are coming. Better data capturing, which predicts the crop health and crop volume, enables farmers to secure better price and also secure to basically guarantee or claim insurance coverage in case of damage. So this is about the agriculture. Now, even if we produce more, it's not going to tackle the global fruit crisis that we are heading towards unless we reduce the wastage in the food cycle. And that's the angle that I'm going to cover. And there's several interesting solution here. One of them is a London-based startup called Winnow, which basically comes up with a smart bin. It has a AI camera, I mean it has a camera overhead, and then it comes with a smart meter. So as in the restaurants, it's basically for commercial kitchens where you can, the camera can catch and identify the food, the quantity of food, the type of food, and the cost of it. It can, basically the cost will be computed. It will show that as in real time, as it has been, this will help the restaurant manager and say chefs there to come up with the right portion and right recipe for the products, basically for the food items because things which are not like, say, suppose capsicum in the pasta and kids are not eating it. You may introduce something which kids are, so you may not know because people have already ordered, but unless you look at how much has gone in the trash, you may not know, and you may not be able to control that. So that's one side of it. The other side is demand forecast, retail. How many times you have gone to a retail shop and said, oh, this mango is too ripe and I'm not going to take it. That just goes to the trash. That's another thing which can be changed basically. So one thing is demand forecasting on the retail side and second is the quality of the product that you are putting in your shelf. So for that, there are several actually products and that's why I haven't put any name here. The third thing comes in our own kitchen which is like what to buy because it happens with us all the time when we go to a grocery shop where we are, say, we do not know whether back in the fridge we have tomatoes or not and then we end up buying, say, tomato which we do not need. So for that, there is a smart fridge camera which looks at your consumption patterns. It also looks at when you bought the food and then it will predict, it will, based on the order in which you bought the food, it will recommend you to consume that first. So this is, and once, I mean, it's yet to, I mean, Samsung and all have their own smart fridges but then this one is a smart fridge camera which is available in UK. This is much cheaper and this comes with a mobile app which can, and once these technologies come to, say, Indian market or in any market, I mean, if it's commoditized, this can be linked to, say, your grocery app with your account and then you can create your buyer list, your shopping list, et cetera. So all these applications will come. I mean, right now it's not available for sure. Now, so we have looked at the consumption side, how it is, AI is helping in, say, cutting down the waste but then comes the food cycle, right? The entire supply chain. So from farm to fork, the food wastage is huge. It's almost like one third of the food that we produce gets wasted. And to cut down that, there is a startup called the Transparent Path, which is a Seattle-based startup which creates, which uses IoT, blockchain and AI for creating a completely transparent supply chain system where you can track the food item real time. And this is, this Transparent Path, actually my co-speaker, Badri Narayan is in the board and he will, he can give more insights about it. IBM also aims to build similar food supply chain in India and they aim to create a zero waste food supply chain in India by using similar concept. It's an integrated platform which they are trying to build where everybody, basically all the participant or all the stakeholders are connected. In the supply chain platform, it's retailer, it's the farmer, how much to produce, how much to buy, how much to, how much to supply, where to supply. All these will be in a single platform using the blockchain technology. So that's going to be, probably can cut down the waste in Indian food supply chain system. So now we have looked at, say, so we are growing more and we are, say, we are also cutting down the wasteage. There's surplus food and still, and then government is coming up with a lot of food subsidy and entitlements, et cetera, for poor. What comes in between? What still, you know, there is, WorldWag feels that corruption is the biggest challenge, especially in developing countries which deprives poor of their entitlements. And for that, I have just put together this proposed framework, there's no solution like this, but this is based on a framework or a model which was deployed, which was actually tried by an NGO called Cuts International. It's Consumer Unity Trust Society. They basically came up with a consortium of civil society officers, which actually used to go to villages and they used to create awareness among villagers about the different, say, entitlements and their options in case they are asked to bribe or in case they are deprived of their entitlements. So at that time itself, RTI Act came and so it was mainly to create awareness about RTI and how to enable poor citizens to use RTI to get what they need. So what is proposed here is quite like a chatbot, which, like a customer help desk kind of system, something which has to be trained for a particular specific set of problem which applies to that particular, say, domain or that particular geographical area. The NLP, as you know, there are many startups in India. LiveAI, which was bought by Flipkart recently, they support NLP conversational interface in some 10 Indian languages. So similar concept has to be applied, all those typical components of chatbot has to be there, NLP, speech-to-text, search, et cetera. What is proposed here is when you are doing the interactive conversation, using that, whether you can find the genuinity of the problem, severity of the problem and categorize the problem. And once the problem is categorized, can it be used to take some action? And their corruption is also of many different types. And one of them is, of course, employer place, workplace corruption, which is especially among poor, they are not given their, I mean, commission is cut sometime in their own salary, et cetera. So such things can be reported, and if at all the system can take the employer, if it also happens in some of the big MNCs. And if that can be used to, say, report to their Grisvam sale, then reporting in media, if there are multiple complaints about the same person or about the same officer, complain to higher authority, and then connect to subscribed, so there are many lawyers who work in this field and who provide, say, free assistant to fight corruption, so to poor people. But then bringing them all on a single platform and connecting them to, say, the poor folks, and then also informing them, creating awareness about the entitlements that government launches, filing RTI for them. So this is one of them. Now we have handled this corruption also, and the third angle could be there are still kids. So we have, like poor also are getting their entitlement, but there is one section which is kids on the street who are basically meant to starve, because, and they basically go hungry because they are forced to starve and they're forced into begging. So that's what is the next framework about, this is about how to save, can AI curb childhood trafficking and save children from starvation and begging. And here it's, so what I know is, it's not that none of this exists today, some of this is available and in different forms, they're different, if you know, Kalash Satyarthi, he has a Bajpan Bachao, which has a facial recognition app, and then Intel is working with a company called, with a NGO called Thorn in US, which is again, working on the facial recognition part, but the first model which we are talking about here is not available, and that's what we are actively working on, it's like a few trials, I mean, we haven't had any success, but what it proposes is to look at the kids on the street and kids who are begging and classify them, whether they are actually a kid who is in, who is in a wrongful position, so who is with, say, a beggar, any kid who is with a beggar, that means the kid is being used to beg because they are meant to look vulnerable, so that's like, and typically people feel more, you know, compassionate towards them and end up giving money to them. That picture is a picture which I myself took at BTM Signal, that baby was drugged and it was sleeping in the like real bad noise and sun, so that's just not possible for a kid to sleep that way when she's like dragged through, you know, swift movement and all. So that was one picture that I took and the third one is a begging kid. So they are like different categories and if we can bring them all together, put them, of course, this has to be from, so again, there are another set of apps which exist like Helping Faceless is an app which does one part of it, if you upload the picture, then it will try to match with a missing kid, that's what it tries to do, but there is no system like this which is comprehensive enough to, you know, identify kids on the street and match it with the missing kid and of course, when it comes to facial recognition system, there are a lot of challenges there, one of them is that it has age invariant application but it does not have growth invariant, in the sense the kid facial features completely change, it's not like how you age, aging is a different process, growth is a different process and it is also possible that you do not, I mean, the time passes that you may not even have the picture, right? So for that, we can try say elder sibling, parents old, you know, parents old picture, et cetera, so that's like, that can be added as additional features, et cetera, to create a set and of course, you know, comparing the missing versus reported can be a very, very time-taking exercise, I mean, you cannot just do it, so probably applying some temporal filtering to see and special, basically, when a kid is, say, when a kid is abducted, there is a high chances that the kid will be trafficked through, say, from outside the city within seven days, next seven days. So if you can send it to the feed of people who are taking train from there, again, capture from the, take the CCTVs from the nearby stations, from the trains that have left from there, if you can apply such filters and then match it, there's a high chances of finding some match there. So that's the idea and this is one part of it we are working and of course, this is a very huge problem and anybody can pick this problem and look at it, look at solving it. There is a high court order on Delhi government to procure the software already, so there is definitely a potential there. And coming to, yeah, the next topic, my co-speaker is going to take us through how economic policies are going to enable wider adoption of these innovations. Thanks. Good afternoon, everyone. My name is Badri Narayanan, Gopala Krishnan and I'm from Seattle. I have a company called Infinitsum Modeling, which is present across the world in four countries and I'm also with the University of Washington Seattle and I'm an economist, so the remaining few minutes that I'm going to talk about would involve a combination of economic policy and some research that we have done on various aspects of poverty and how policy can address that and to some extent we'll talk about some of the technologies that we discussed before. So typically traditionally the economists use data from economic surveys. So in India we have a national sample survey, so the picture that you see here, which is pretty widely respected across the world as one of the most trusted surveys among all the developing countries in the world. But still there are a lot of gaps in the data and I can say as an economist, as a researcher, I've used these data sets a lot and there are a lot of difficulties in terms of missing data and lack of coverage of population and so on. And so there are two aspects here. One is to use these kind of data sets for analyzing things like food security, water scarcity, employment and so on. So there are two kind of ideas here. One is using these data sets so you can take that these are the best available data sets and you can still use much better state of the art methods like those in artificial intelligence and vision learning and do a lot of analysis. And second, even better way of thinking is like what Usha was talking about, use deep learning kind of methods to create the data using technology rather than going for surveys, which are manual and error prone and so on. So those are the two aspects and for the next few slides I'll be mostly talking about how we have been using these existing data sets and doing some research on this. So then you see organizations like World Bank and our company and many others in academia and so on. We have been collaborating with organizations like World Bank to understand what are the economic implications of poverty, food security, lack of food security, water scarcity and so on. So like in the first slide we talked about the Nobel laureate Schultz he mentioned, if you don't know enough about poor people then you don't know enough economics. So in that sense to understand the sufferings of poor people in economic terms also matters a lot. So we did a large project with the World Bank to estimate the economic impact of water scarcity in the next few years. And of course our analysis was limited by the official data sources and so on. And but it was still a good starting point and we have now been working on using this disruptive technology, some of the disruptive technologies to solve the problems like food security and water scarcity and so on. So for this we collaborate among economists, technologists, data scientists, people who are working on these other disruptive technologies and so on. So this collaborative work in which you first identify these technologies and then you think about the cost effectiveness of different technologies because we can come up with a lot of nice ideas but they should be effective and they should be better than the previous methods. And so and also the company that we discussed, Transparent Path, there have been many companies like that which have come up with some practical solutions which are also viable commercially. So taking the particular case of Transparent Path, what we have been doing is to come up with a score using data science for every, like we already did a pilot with apples, Michigan apples. So you have a QR code on the apple and if you scan, you can see the entire supply chain, where it came from, what was the temperature and humidity when it was formed and what were the chemicals involved. And even if the chemicals are not involved there, were there any chemicals in the neighboring fields and so on, so something that might have cropped in. So a lot of these details. So basically it's all about utmost transparency and we give a score based on the extent of transparency and this is all work in progress. So we have sensors in the food cultivation processes and the entire supply chain which are coming with IoT and you have the data in the cloud and then you do some analytics on top of it. So these are some examples. So now I'll move on to some of the traditional kind of approaches, how we have been addressing poverty as a policy researcher or policy maker, policy analyst and so on. So I'm just giving some examples and I'll quickly go through them. So one aspect is analyzing the effectiveness of policies. So policies are already there. How good have they been performing? And then you can calibrate, you can improve it, improve the aspects which are not working and introduce the aspects that may work better and so on. So this first paper we looked at the US agricultural subsidies and see whether they have actually led to an improvement of farmers in terms of their incomes and so on. And we found a lot of intricate findings which said a lot about the details of the policies and how they can be modified to be more effective. And the second paper is about there are, so there are two types of policies. Some policies are aimed at eradicating poverty, they're anti-poverty policies and there are other policies which are, in general, they're good for economic growth. So trade facilitation is one of them. Improving the infrastructure, better ports and so on. So how would that affect poverty? So that is some analysis we did and all these are data intensive analysis. And then we have done some work on sustainable development goals and how are they affected by other economic policies, particularly trade. And I just talked about the water scarcity, impact of water scarcity on the economy and the economic growth on water scarcity and so on. And then we've done some work with FAO, Food and Agricultural Organization. They are interested in what are the ways to encourage farmers' income security for them and that leading to food security. So this is just summarizing what we discussed. One additional point I would mention here is they're having these traditional analysis that are done so far, but now we are coming up with some very new policies like doubling farmers' income, universal basic income and so on. And also direct benefit transfers. So in this context and coupled with the availability of data using other and digitization of transactions and so on, there is a huge role of artificial intelligence, the new algorithms and so on. So I just wanted to mention that as an additional point. So I just wanted to explain a little bit of, to just give you a better picture of how this kind of analysis are done so far and what role AI can play. Just one study I wanted to elaborate a little bit. So many of you may be familiar with the National Food Security Act of 2013, which is regarded as the largest social policy experiment in the world. And that basically involves giving subsidy on rice and wheat and certain grains for different sections of the population. Mostly rural power, urban power and so on. So we did an analysis to see, we did this way back when the policy was just announced and implemented. Of course, there was not much data after the policy. So it was an ex ante or analysis that was done with a kind of predictive kind of intention. So we looked at what is the impact of this policy on different types of households and consumption patterns and so on. And we found that if you give subsidies linked to specific like rice subsidy or wheat subsidy and so on, it's going to change the consumption patterns of people. So if people are consuming, like Usha was talking about sorghum and there are some healthy grains like millets and so on, they may shift their consumption towards rice and wheat because they are cheap. So the alternative is to give income transfer. So you just give them income and subsidy and then let them decide on how to spend that most. And then in this case, our analysis was based on survey and the traditional national data sets. We could have done much better job if we had some AML techniques to collect the data and so on. And there are three, I'll just take a few more seconds, I'm almost done. So there are three different stages in policy. First is planning, second is fundraising and third is project implementation. In all these three, a large number of economic studies are done for planning. So before the policies happen and for fundraising, we have done some work on that to target the kind of donors and so on and implementation and effectiveness. I gave you some examples on that. So the challenges are as data sense community, we all agree that there is a trade off between getting deep into efficiency and data-oriented decision-making and the story behind it, the insights and story. So this correlation versus causality is a classic example of that. So how are we going to tackle these issues is something that's still there. So my personal feeling or recommendation on this is that we all should work together. There are subject matter experts, domain specialists and so on and there are data scientists. So we all should work together and tackle these problems. So I'll just keep the slide live and end. So basically I wanted to just come up with some list of some organizations. It's not by no way, it's an exhaustive list. It's just a very partial list and which just came to my mind. And there are many such organizations where we as data scientists can collaborate, submit our proposals and so on. So thank you very much. This is some references and. Okay, so we have time only for one question. So I have two questions. It's two parts. One is that how is the acceptance or adoption of these kind of solutions in the context of India? And the second thing is how does the Indian, how does Indian ministries or Indian government help in enhancing the data ingestion process for such kind of solutions? Yeah, very nice question. Thanks for the question. In India, there has been increasing interest in these new technologies. If you look at the NITI-IO, the new counterpart of Planning Commission, not new anymore, but so you can see a lot of reports, research reports and documentations, which actually talk about the power of artificial intelligence in countering poverty and various things, blockchain. And actually blockchain is not that much in the radar of the policy makers, but I would say that irrespective of the political leadership, I think there has been an increasing tendency towards looking for technological solutions for various problems that you're facing. And that is where I was interested in showing these things. Like particularly in the context of India, it's like other innovation mission and other schemes. NITI-IO has been a very focal agency in this. And other ministries are also involved, but most of these things happen through NITI-IO. So I would strongly recommend all of you to keep looking at the website of NITI-IO and follow the developments there. Yeah, so my answer is positive. Thanks. If you have more questions, you can connect offline with Badri. Thanks.