 Our next presentation is from Sumya Ranjan and Shravan Kumar talking to the point of AI for the marine environment biodiversity conservation and social food. Hello, welcome to the session on AI for the marine environment biodiversity conservation and social good. My name is Shravan. I'm the director of clients at Susat Ramana. We are a data and AI company with presence in five locations across the world. So our solutions we intend to bring data as insightful stories. And as part of that, we have a huge body of work in the EHE sector, particularly in the underwater and marine conservation areas. Our teams have been constantly building cutting edge products that can help conservationists, managers of fisheries, aquaculture professionals and GIS officers in various organizations. I'm going to give you a quick glimpse of the solution. And our goal is to make tools and methodologies like this and technology like AI available to people in organizations that are working to solve these problems that help create a better sustainable blue planet. And in that pursuit, you work with the likes of World Bank, Microsoft Philanthropy and AI for Earth and Bill and Melinda Gates Foundation to name a few. So why is creating a sustainable and blue planet important? As it turns out, fishes top the list of animals that have the highest number of species that are endangered as per the IOC and Red List. The 3000 odd that are on top of this list form a part of 35,000 fish species that have ever been registered in the tree of the plan, which makes almost one out of 10 fish species vulnerable to extinction today. If that was not enough, one third of global fish stocks today are over exploited, which means they have produced using unsustainable methods, which means the margins could be lower and adding to that problem is illegal fishing that is rampant across the world. How can AI and technology solve these problems? At Grammar we believe that, you know, custom data solutions for underwater AI can help answer some of these questions. I'll start with our AI as a service for underwater species detection, which helps, which has helped in the past impact conservation groups, foundations, zoo systems, you know, identify and classify species and maintain a record of their sightings so that they, you know, are alarmed about anything that is an aberration in their accounts and sightings. And using computer vision, we have been able to achieve that and institutionalize that in multiple organizations. Using computer vision again, we've been able to estimate biomass in commercial species where there are huge tanks of, you know, fishes and there is a cycle of growth that the fishes undergo. So you have to also, you know, monitor the environmental parameters like dissolved oxygen, pH levels of the water, and so on and so forth. So biomass estimation through the computer vision, plus the, you know, environmental parameter monitoring, helps them, helps the commercial fisheries have better visibility of their produce, and also do so sustainably. Last but not least, we also have a school of GIS solution for marine management, which can help track water quality in the ocean, track ships and vessels that are not supposed to be there, and are taking part in illegal fishing. This can help ecological researchers and even commercial fisheries from, you know, falling prey to illegal fishing. So I think these three solutions can answer some of the problems we spoke about, but I quickly, you know, jump into a quick glimpse of what our solutions look like. So here in this slide, I have a couple of videos, which when played, show you that a certain fish is parking through a tunnel in which a camera trap is fixed. Whenever the fish pass through, the camera starts taking the video. And if you see here, as soon as the fish passes through the AI model is able to detect that it is a certain species. And even a part of the fishes, you know, head is enough for the AI to detect that it is a certain species. Showing you another example of a smaller fish, but a similar tunnel. Here, not only the head, even the tail is enough for the AI to pick up that, you know, this is the classification. That has to be done for that particular fish. Right. And this has immensely helped the river conservation agency that you've worked with in the past. And, you know, their problem was to actually, you know, do this detection to, you know, binge watching all these videos and having their, you know, biologists watch these videos and tag the fishes with their species. Right. It requires two things, a lot of time. And secondly, the biologist knowledge or domain expertise, which helps them identify your species as it is. So we trained the AI to acquire that knowledge, and you know, tag species automatically, which helps in reclaiming all the resource time, otherwise spent in just watching videos. Now, you know, for the person who does not understand this technology, you packaged it all in, you know, a web app or an app of sorts, which can help bring out these insights or the number of sightings, the trends, if something is going too much up and down. They can be quickly notified of any aberration in the environment, which enables them to, you know, monitor this very closely. And hence, you know, draw attention to something that might go wrong. This can help a lot of conservation agencies. There's one kind of solution. And then when you bubble this up with multiple cameras and multiple, you know, locations, you can also do biomass estimation of a fish or a school of fishes. So the same thing was applied when we work with commercial fishery in Indonesia, who wanted to track in each of their, you know, fish tanks, which are huge, you know, 18 meter to 15 meter and of size, where they had these fishes, and they used to grow the fishes over a period of time. And, you know, before the solution, they didn't have any visibility of their yield, of their produce, what they're going to get, and they had no control whatsoever. We helped them achieve this by, you know, just installing a couple of cameras overhead and underwater, and a bunch of sensors that, you know, measured the dissolved oxygen, pH levels of the water, and other things that help inform the conditions in the water. And hence, correlating that with the growth of the fish. Now, when you're able to do both, you can correlate this with an AI model with precision feeding analyzer. What that means is you're able to say that based on the, you know, size of the fish or the stage of the fish growth, they're able to feed the, you know, fishes accordingly to avoid underfeeding or overfeeding. And this enables you to have better yield, better control on your environmental parameters at any given time. If there is a parameter that is going out of, you know, the ballpark, they can quickly be notified and necessary action or intervention can be taken by on ground personnel. Now, summing up all of this, the outcomes of this solution was one resource time gain, like I said, highly skilled people watching videos is no good for anyone. So we reclaimed that resource time by almost 98%. We reduced the manual effort as you, you know, saw the IOT cameras or the, you know, connected cameras to all the data collection and tagging and the AI used to do the tagging. No manual effort, all the manual effort was brought down to zero. And then in the case of biomass estimation and the commercial fishery use case, there was a huge saving over a period of three months, we observed 20% of cost savings on the feeding alone. Right. And then the bottom line, also getting impacted was kind of impact that we were able to create through the solutions. There are some success stories, but our solutions come in all forms and shape a similar technology but a different favor was used to count penguins and talk to take off, you know, classify and detect species in, you know, a natural park. We were also able to use this to avoid elephant human conflict to just give you a few examples. At last, I would like to leave a message for everyone that's, you know, looking at this is again reiterating our vision of sharing these tools and methodologies to people in organizations that are otherwise not equipped or not invested into technology. So we want to, you know, get to a stage where this AI can run on even mobile phones. And we want to build a knowledge for a database that can act as a base for others to come and build on top of it. So it's a huge shout out to everyone out there to, you know, collaborate with us. And for that, you can get in touch with us on on the links that you see that I'm a dot com. Or you can tweet to us at parameter or on my personal ID at show and legacy on the screen. Thank you, everyone for the opportunity and we hope we'll cross paths very soon. Thank you so much. Thank you for sharing show run as someone who's spent a year or so feeding fish in salmon pens in northern Scotland and has to stay up overnight with infrared glasses on to camp penguins coming up beaches from to their nest boxes. It sounds like you've got some great breakthroughs there. So thank you very much. Shevron spoke to fish species as having high extinction rate. Now I would just like to urge everyone to clear up a number of misconceptions about what the IUCN red list characterizations mean and suggest that reading the criteria will help you see why marine fish are likely a group with the lowest extinction rates of animals. But let's get on with the question and the question is going to Sumia part of show runs team. And this is really about talking about partnerships between agriculture and fisheries and it's obvious that some of the greatest growths in understanding fish stocks has come from agriculture because obviously we get a much closer view on how fish develop and so on. I'm just wondering now that you've pointed out that you're developing some of your tools for agriculture and these are going to likely to cross over to fisheries. Just give us some idea on how long it's taken you to get to where you are today where you're already talking about spreading this technology across other animal groups how long is your team been active. Thank you. Thanks Kim and thanks everyone for having us here. So we started our AI journey I think four years back with Microsoft with their AI for the initiative. So initially we started out detecting species on land like different animal species. And then we moved into drone imagery where we used to track elephants in forest and make sure there are as less human conflicts as possible and slowly moved into fisheries. So the models that we have built in housing grammar they are agnostic to any particular animal or fish in place like they would work out of the box if you have label data available. So what we do as a technology company is we have our algorithms in place which are based on state of the art models that are currently out there which do the work perfectly. When we on board a profit or a nonprofit towards right what we need from them is data most of the time they do have data it's not in a labeled format. We either help them in the labeling process or we like like we generally employ third party who comes in and labels the data which requires some expertise in that particular domain. So to answer in short it's been three and four three to four years we have been working this piece. And now we have a better visibility of how we're heading and how the industry is headed like even when I was going through the agenda of the talks that you have lined up today right. Like there's a lot of overlap in the work that we are doing even the previous speaker when he was speaking about fish ID and the work that we have done I see a lot of overlap there and that's something that that's happening across the world just that we are not connected. So if you could build a database of labeled fish and a database of good models that are out there, which are working out of the box I think that that would help everyone. Thank you very much. Yeah, I'm really interested in your work it's fantastic presentation thanks to the whole team for putting it together. I'm quite interested in just the cost of investment to do what you've done in that work and and how you came to how you came to do it and also your future trajectory in terms of reaching out to people that might not have that sort of investment to to implement such technologies you know, make it more low cost more more accessible for small scale aquaculture, for example. Like I said, like we have we have a very small team in place in grammar. It's called the grammar labs and I had the team. So it's a small team like we have around 10 odd people working mostly in computer vision space. Over the past years we have developed enough expertise and we have spent countless hours building the models training them out using all the GPUs and all which is in fact impacting them on right. So we don't want people to repeat that process that's why it's not just the for profit organization we work with we work with a lot of nonprofits right so we don't expect them to like start the journey or repeat the process that we have done so far. We want people to build on top of us right like when we say that we have computer vision models, generally we have a pipeline in place and it's an active learning pipeline. We won't give you one single model right we'll give you a pipeline, which would do let's say 70% or 75% of the work that you're currently doing. And it might be anything it might be fish hiding it might be tracking it might be classification it might be biomass estimation right. Our pipeline is designed to like to be flexible enough to adapt to these kind of tasks and it would start with a decent amount of accuracy, and then you have a active learning pipeline in place which will then help you out to re label your data correct the mistakes that the model is currently doing and with time it evolves along with you. So it's like general software engineering you don't have to build from scratch, you sit on soldiers of gents and then you build on top of it. And that's what we are providing we have something called Grammix. It's an open platform it's on GitHub anyone can access it to start like using the models for their own data, and then build on top of it reach out to us if they need any help there.