Building an AI platform with a deep tech approach

‍

Transcript

Vishal: Welcome, everyone. I'm your host, Vishal Katariya from Ankur Capital. And today we're continuing our exploration of the world of advanced computation and AI.  

In case you missed it, we recently launched the India Deep Science Tech Report, a first of a kind comprehensive dive into the transformative world of deep science technology in India. The report unveils the cutting edge breakthroughs and investment trends that are shaping the future of Critical industries like healthcare, food, and energy,  and the journeys that ventures embark on to build hugely disruptive products to solve our most pressing and urgent problems.  Combining qualitative insights from our decade of investing in the space, as well as quantitative investment data analysis, the report is a treasure trove of insights and inspiration. Find the link in the show notes.

With us today, we have Sujit, who is Chief Customer and Marketing Officer at Cropin, as well as Praveen, who's the head of Cropin AI Labs. Founded in 2010, Cropin is a global Agtech leader enabling intelligent agriculture. Cropin Cloud, the world’s first industry cloud for agriculture, empowers stakeholders to make informed decisions that enhance farming efficiency, productivity, and sustainability through digital technologies and predictive intelligence. Cropin is spearheading the ag-intelligence movement through strategic partnerships over 250 B2B customers. Cropin has digitized 30 million acres of farmlands, and positively impacted over 7 million farmers worldwide. Their crop knowledge graph spans 500 crops and 10,000 varieties across 103 countries. This is what powers the Cropin cloud in providing predictive intelligence for over 200 million acres of farmland globally

We've been an early and proud supporter of Cropin for a decade now, and seen their offerings scale up in value from a digitization platform to a predictive intelligence platform.  This TechSprouts podcast episode is slightly off our beaten track. Where we're going to dive into Cropin's journey of adopting a deep tech approach to building Cropin Cloud on top of their digitization services.

We're very excited to have both of you on this podcast, Sujit and Praveen, and eager to learn from you over the next 30, 40 minutes about the journey in creating an AI based platform, the kind of  USP or domain expertise it takes to do that both on the computational front, as well as on the domain specific knowledge. And then at the end of the podcast, I'd like to zoom out and talk more about AI in general, how you think about protecting AI models and how do you keep up with the rapidly evolving world in AI.

So my first question to both of you is just to set context and open the floor. I would like to know a bit about your journeys to get to where you are, both at Cropin, and in a bit more detail, what skills have helped you combine the cutting edge tech, the subject matter expertise, and the business experience to create the Cropin cloud. And if I may, I would also like to kind of ask what fraction of these three was needed to do that. So Sujit, you can start.

‍

Sujit: Thank you Vishal for having us here, super excited to share our perspectives from a Cropin lens, both in terms of our journey, and also address some of the things that are very fundamental in the way solutions like Cropin are being built out and taken to market.

So, glad to share those perspectives. So to answer the first one, Vishal, I think if you look at my own personal journey, it's been about taking different products and services in the technology space to market in, into markets that are in different stages of maturity. As well as products that have been in different stages of maturity.

And what, what I've really learned along the way is as long as you keep your focus purely on the customer or the user that you are trying to solve problems for or solve for opportunities that those customers have, you will get your journey right. And that laser sharp focus is pretty much what's the fundamental, skill set that I have gathered and I've put to use even during my journey at Cropin.

Ever since from day one, when I joined at Cropin, understanding the market was super critical. And one of the key insights that we gathered was the fact that the market is still in very early stages of evolution, which meant that it's not really mature to be very clear about what it wants, and how would we want to solve it.

Also, there were challenges on change management, et cetera. All of these insights could be gathered only because of the past experience, as far as my journey is concerned, of understanding these nuances of product maturity and market maturity. And I think that's been very critical, at least from my perspective, of what I could bring onto the table as part of my journey with Cropin and as we have navigated the evolution and iterations that we have made in the last few years.

That’s a bit about whatmy journey has been, Vishal. Praveen, over to you.

‍

Praveen: Going from a researcher perspective and coming to a startup, it's been quite a bit of journey for the past three to enough years, um, since I joined, Cropin, and it's been a wonderful journey so far.

Regarding the, you know, the cutting edge versus SME expertise and business, what I would say is that, we have been fortunate enough to have a lot of interactions with many of our clients who are already experts. And we have within our team also quite a bit of expertise. 

In  from after all years, working with a variety of crops, across different geographies and everything. So it's been wonderful, uh, learning from all of that experience and putting it all together. And, you know, so the, the key, the key element is how do you combine the SME experience  with the latest solutions that's coming in AI and integrate together, them together in a way that it can actually serve the customer and it's useful for them, right?

It's not really about creating different things together, but mostly focusing, really on the usability of it and the direct adoptability of it, right? In the business side that I’ve been a little bit, I would say learning much more from this particular experience and what I have to say is that it's always about choosing between two important parameters.

I would say one is complexity versus simplicity. So the most you know, simplest models often are the ones which I found like really solving business problems. And there are some. cases where we really have to deploy and engage complex models, right? So to see how that actually we have to do that and kind of like making a trade off between that the complex versus simple model cost versus investment right.

And scaling, and all of these are not purely technological problems, but are related to a lot of business problems, right? 

‍

Vishal: Thank you, Sujit. Thanks, Praveen, super interesting. And I think the first thing I thought of when Praveen, you've said it's an interesting journey from researcher to startup.For me, it's also an interesting journey, a researcher to an investor. And the other takeaway I have is that it's important to keep the customer and customer in mind, not only from a business perspective, like you mentioned, Sujit, but also from a subject matter experience perspective, because both of these things are what are guiding the product. And I think the last thing you said Praveen, was that it seemed to me like what you're doing has to be more than the sum of the parts. It's not just being a middleman who's combining data and model, but a lot more that the subject matter experience is kind of unlocking.

So the next broad theme that I wanted to kind of move towards is. What's the journey like to create a platform like Cropin Cloud? Is there any, you know, fundamental breakthrough or a USP that enables spinning off multiple products  from this platform? Did that happen at the design stage or did that kind of happen retrospectively when you realized, hey, we actually have a platform technology?

To add some nuance to this, what's the kind of considerations, decisions that took place as you started from algorithms and ended up at this product.

‍

Sujit: Cropin started way back in 2010. And this was a time period where Ag-tech was not even a word. Most debilitating problem that the sector was facing was just the lack of information or real time information from the field or the farm and Cropin really started building digital solutions which was primarily mobile and web that could help collect and aggregate data in a structured manner from the farm and the field.

So that sitting centrally at some place is businesses, governments and government agencies or programs could understand what's going on. And  instead that help progress towards the right objective that they're set for their their business or programs. And while we started there, the immediate next step that happened as part of the evolution was these customers who were adopting our digital solutions started asking us a question around, Hey, you're helping me bring all of this data together via your applications, but I want to be able to do more with this data. I want to be able to triangulate information.

I want to be able to understand there are other nuances that I'm probably missing out and that’s how the journey around, how can I help my customers manage this data better, which was the data management layer came to play. And while that journey was, was evolving is when we also took the futuristic bet on the fact that, hey, we don't want to just stop at helping customers manage and understand more from the data that they're collecting via our solutions or by just bringing in data from other sources.

What we also want to enable them with forecasting capabilities of predictive intelligence capabilities. And that's where came about the origination of the Cropin AI Labs team, which is the team that Praveen heads up which is a, which is a group of scientists from the agronomy space, data science space, AI/ML experts, cloud experts, as well as other observation sciences, where the primary goal is, can I use a remote sensing based data sources which could be satellite, weather data, et cetera, and create models that can help me provide predictive intelligence solutions solving for core challenges that the agriculture sector faces and that gave birth of the third layer of our solution, which was intelligence. So if you see this journey, it's been a constant evolution of building solutions that the customer or the user really needed.  And then eventually when we got all these three components together, we realized that we had got the perfect mix of the building blocks that were required to actually create a platform and that's how Cropin cloud came into being.

‍

But I will, I will share more around the Cropin cloud platform journey, what, what went behind it, et cetera, as we, as we move forward in this discussion, but that's a little bit of context to understand how did we evolve to the Cropin cloud platform stage. Praveen, is there anything you would like to add? 

‍

Praveen: Yeah. Yeah. Thanks, Sujeet. I think you have covered a lot on what I wanted to also talk about. And I think there is one key element which you also mentioned is solving for some key problems early on is very, very important. Had we, like, started off from with some of the bigger problems then you know, we would have probably lost down on some of the foundational elements, which was key you know, before.

Even the you know, the intelligence comes in, right. And I think that getting that right at early stage was very, very important. And the other thing is you know, is that in this journey also we had constant interactions with clients and researched with the SMEs, and we are always looking at not building systems for one particular crop or one geography. It was always looking at scale, scalability,  right? And that was a key element as well. So at the back of our mind, when we are building our models, so the element of scale is important to have as well, right?

‍

Vishal: Thanks again, Sujeet and Praveen. Praveen, I think I, the only follow on I have to that is since we were talking about kind of is that our underlying USP,  my conclusion was that it's coming from the experience from digitizing the data and, and building up the subject matter expertise.

‍

Praveen: So would it be fair to say that.  That's where the cloud’s differentiation is coming from. And so what has happened is when we are building these systems you know, over a period of time, we build it in such a way that, that's why I was talking about scalability and that's, you know, the, you know offering that we have is that you know when we are bringing, scaling it up, then automatically we're looking at enabling it on the cloud platform.

And that means like the ability for us to onboard people and with minimal amount of effort and time. So that it's, it's, it's usable for a maximum number of people. Right. So that was a key element that we focused on. Yeah, absolutely, Vishal.

‍

Vishal: The other the next question I wanted to get into is, you know related to the way our TechSprouts podcast has been structured, even over the last two years, where usually we have interviewed earlier state startups who have been at earlier TRL levels and have been at a prototype or even pilot stage of their foundational product. 

Cropin Cloud, on the other hand, is something, as Sujeet mentioned, that came about as an organic. layer that was built on top of the foundational work Cropin did, and it came at a stage where Cropin was a scaled company, not, not a, you know, prototyping in a lab or a lab basement stage. So one of the things that we thought was really interesting or interesting to bring out in this podcast episode is the difference in creating such a IP-led or deep tech platform in a mature company, vis a vis how it would have been in an earlier company. I think already we have some, some insight into that, that once you've established yourself in the space and have access to data, have customers who are willing to engage with you, that makes it easier. Sujeet, over to you to kind of build on that and elucidate a bit more on that and the way this was different than how it would have been at an earlier point.

‍

Sujeet: Ourselves in making this pivot of moving from a bunch of point solutions that we had built out that were already seeing adoption and scale with customers was the, the insights that we had from the market because we were really, really observing closely what our customers were really struggling with and some of the key insights that we gathered.

Through that journey of making the decision was point number one. There was just one too many or two thousands of point solutions out there in the market and in the active domain, which was making it next to impossible for customers to make decisions faster on which solutions do I go ahead with and will it really solve my problem?

The ability to democratize data was becoming challenging because they were all different solutions being stitched together, some that they built themselves, some that they acquired, some that they procured and some that they were hoping they could evaluate and bring into their portfolio.

And so we realized that was the biggest problem that customers were struggling with. And like I said, technology adoption in agriculture was still in, or is still in actually early stages of evolution and adoption, which meant that they would want to make bet on partners who could be long term partners for them in their digital transformation.

And that was the primary insight that got us to pivot to the fact that let's bring these point solutions that we have, which was the perfect building blocks to build a platform for solving for the agriculture industry. And the good part is again we learned from our collective experience.

For example, me, Mohit, Rajesh, Praveen, all of us coming from the enterprise space of the business where we clearly understood what enterprise customers need, what do they understand very clearly. And all that we were doing was building a custom solution for the agriculture sector. And hence it became the first industry cloud for agriculture, which is now called Cropin cloud.

So that, that was a journey of decision making and it was primarily driven with insights. On what were the challenges that customers were facing and the moment we took this to market we saw huge resonance. We saw customers Looking at us with a very different lens even when they were evaluating us they were no more comparing us with individual point solutions, but for really challenging us on what would we really solve for them in the longer term.

The decisions really became more strategic with that clearly showed even even in an average deal sizes our revenue roadmap, our engagement, maturity with these customers all of that clearly started telling us that we we had made the right choice and decision and that's been the path that we've been taking The other bit of evolution that that we also committed to when we were taking the Cropin cloud journey was the fact that we are going to pivot ourselves to be an ecosystem platform, which meant that we have to be an open platform where other solutions can, can easily integrate into and integrate out into, because we were very clear that we won't be able to solve all the problems that the sector faces, and we want to ensure that we can unlock time to value for customers and help them accelerate it and that's where came the whole other component of our our commitment of ensuring that we will make it very very easy for customers to really integrate and unlock value from their investments that they've already made and investments that they would probably want to make with other solution providers for areas that we don't solve for, so that's really been the the inside story around our whole evolution journey and the decision making process. 

‍

Vishal: I think  it's been really inspiring to hear what, you know, Praveen and you have been saying about how the experience that Cropin has had has enabled the cloud to solve for very nuanced problems and very specific customer requirements rather than going after problems in a more blunt way. And I think, that's to me, that's one of the key differences that's coming from developing this kind of deep tech solution from a company that has this, as you said, collective experience when it comes to enterprise customers, institutional knowledge when it comes to agriculture, and yeah, I think that's, that's really interesting, something that we haven't covered in this podcast. Thanks a lot for those insights. 

Praveen my next question is to you, or rather even my next two questions to you. Another thing that we think about a lot when we are evaluating, working with, or even reading and writing about deep tech innovation is how do you maintain a differentiation or a moat? 

And how does your right to win in a market play out over time? You may be the best today. You may not be, there is no guarantee that you may be the best in two years time. And the added nuance is that AI based or any algorithm based solution is, is not really patentable in the way other material science or biotech innovations are. 

So from a protection and a continued moat perspective, how is Cropin cloud going about this? Is there any particular moat that could be either the data or the way the subject matter expertise have been built, brand new models or is it something else? So would love to hear your thoughts from a technical perspective of how the ring fencing, essentially, of the technology is being laid out at Cropin. 

‍

Praveen: Right. Right. Thanks Vishal. So just also piggybacking on what Sujit had said, like some of the elements that we have been focusing on is like really very tiny innovations in the sector, you know, and so you know when we talk about innovations and protecting it, the fact that you know, some, some organizations can be very, very focused, right?

Like you have agriculture companies focused on the crop of interest that they work on. So they have very, very focused knowledge in that area and these point solutions, which Sujit was talking about was more built around that. And over a period of time, if there are more crops that come in, you have more point solutions.

So that's was how, like, you know systems started increasing and when we are talking about innovations you know, from, from our perspective versus innovation outside you know, when organizations talk about that, then they're talking about innovations from a vertical perspective. You know, we have to look at it from a completely horizontal perspective.

You know, we have to service the entire industry. And that's why I like the innovations are like much more broader, I would say. And in comparison, like to, for example, tech innovations, right? What we have found is it's been very difficult to adopt some of these models or agriculture use cases.

You know, even sometimes the most top-performing models, when we bring that to scale, it actually fails, you know? So we'd had to build these systems ourselves, you know, the models we had to build. And it's over a period of time, with iterations and with the experience that they came to know, like what works, what doesn't work.

Right. And we have been very selective about it. So our way of keeping these models is we keep it as a trade secret for the time being. And as you rightly stated it's difficult to patent these models, right? So where necessary, we have also applied for patents, but much more for protecting our interests, right?

Not as offensive patents. But the key ingredient for us is as you said, it's, it's almost all, almost all of it, right? The very fact. that we can see the variety of data that's coming into the platform helps us into also experimenting with new models, you know, which are probably also developing in the AI space.

So we can try those. And so our current platforms that we have built is modular enough to kind of accommodate that. So our pipeline supports that, so we can do innovations much faster than what we used to do before. And these innovations, like, helped us also to scale up and bring them to production and to customers faster.

So the fact that we have the data moat, the variety, has helped us into experimenting faster and knowing which can fail and which will not fail, right? Because we do have that diversity of data sets. So I would say, diversity is a very, very key factor when we want to test our AI model. And especially that is a very particular problem, which is very key to to agriculture is this out of training datasets distributions, right?

What we call as OOD out of distributions. And those are samples which are like. Outside your training data set, right? And you often you will find that in agriculture what you have trained for and what you find in the wild is very, very different, you know? And so for those extreme cases, those extreme cases actually becomes a majority, you know?

And even in extreme value theory, you know, that it's very, those are the place things which are very difficult to model and to prove for, but the fact that we have the diversity of data set helps us in also testing out our platform and the models. in those extreme cases as well.

So that's a very key ingredient, I would say. 

‍

Vishal: Thanks a lot, Praveen. I mean, clearly  five minutes is not enough to go into all the details of this. There's a lot of theoretical data science. There's a lot of computational stuff that that's going on. And I think what's been, what's happening is the variety and then quantity of data kind of leads you to a flywheel or a virtuous cycle where I think it's enabling you to do better experiments, faster iterations.

And I think the more data you have as well with the data driven way AI is moving right now, I think it also leads to better performance of the product. So I think once you're on that kind of hamster wheel, you know, positive it's, it's really just a matter of things snowballing into better solutions from there.

And I think that also ties into what I was asking the previous broad question is, how does, how do you see this as a platform spinning off new products? And I think this is also a way, what you just said, kind of answers that as well. That the, as it expands, the more data is collected and the more use cases that you can then attend to.

So super, super interesting. And I think as an institution as well, AI is something that we are grappling with, I think as a world as well, that's something that everyone is grappling with. So this nuance that, that you brought out really helps us, and I'm sure the audience as well to understand how an AI based deep-tech solution is being built out or how you would maintain an advantage. 

This brings me to the last point that I wanted to cover on the podcast, which I'm sure everyone has been waiting for, because this is about the extremely zoomed out version of it's a rapidly moving field that is AI, How do you even keep up? From a, especially from a technical perspective, there are these big innovations that happen every five years or so, but there's new papers, new models, new products, almost all the time. 

So  at Cropin cloud, at Cropin AI labs, how are you keeping up with that? And how do you also choose how to participate in that ecosystem? In the sense, you did mention new models are coming up, which you need to adapt to the agricultural context. How are you going about it in a continual way where, how are you assessing, evaluating which new innovations could be something that can make the cloud platform even better than it is today?

‍

Praveen: It's a great question, and I might not have answers to all of them because this is as you rightly said, like you know it's, it's a very, very fastly evolving field. And I feel that even the micro innovations that's right now happening in some of these foundation models and things that, you know you know, within maybe a couple of days, there's new models which are coming and and you know, new ideas, right.

And, and I think it's been one of the key elements, as I pointed out, is in all of this, I think it's, it's very easy to get lost, in fact, to be honest, you know, like we are thankful really to the open source community for sometimes even sharing some of these codes and everything, and we would like to be participant of that as well.

And you know, we've also been internally discussing on how some of our own knowledge  and models can be brought to the outside world and shared openly. So we have been very keen to share our knowledge, of course. And wherever possible, we have been conducting webinars and everything. Sujit himself has been driving that and we do have a distinguished lecture series which we are running where we were inviting people as well. So these are the ways in which we have also been giving back to the community. I feel that's very, very important. And we have also, we want to be the driver of these kind of discussions as well, because often not many people have this opportunity of testing models in the real world, you know, and once we test them, we would like to share that.

You know, see, these are the pitfalls of doing these kind of, building these kind of models, right? And not everyone sees that even some of the foundational models which are being released You know we have already tested some of them and we know where it performs well that it doesn't perform well, right?

So we would like to be also, I may say so be, also guardians of you know, in fact, of people and you know, the roadmap that AI takes towards for agriculture use case, you know, because we don't want people to be lost in this space as well because it's a lot and so we have been you know, keeping ourselves abreast also of the latest developments.

And as I said, we do those kind of testing and see where we need those micro innovations as well. Right. And I think that will be something that we will keep doing, but I just want to leave with one particular thought is agriculture is, is a little bit really complex unlike other systems, right? And there are a lot of moving parts and knowledge and data becomes the key, you know, so we have always been focusing on how do you capture knowledge, you know, how do you codify it how, do you encompass that? And how, then how do you drive models which will actually you know, work on them, right?

So I think that's a very key important factor. You know, unfortunately, the current AI systems has been less talking about that. I would say that in the next year or two years, I would like to see more developments in that area. And I have also been advocating that. 

So, as you know, we do have the AI Center of Excellence being set up in in India, and I've been advocating in that forum as well as part of a committee member that hopefully we can put the entire knowledge together of the world and India and drive our models because that will be a key factor, which will combine with latest developments, like in some of these foundation models or going forward other models, right?

Which will bring some, if I may say so, ringfence to this whole old development, roadmap for it. 

‍

Vishal: Awesome. Thanks a lot Praveen. That's again, very insightful. That the word you use Guardian, that really stuck with me. And I think what you mentioned about the webinars, the sharing knowledge, it ties into what Sujit earlier said as well about integrating point solutions, creating a sort of ecosystem around the agricultural data field and creating an interoperable or an open platform for people to participate. So I think that's it's really an inspiring journey about how. As you've kind of gone up the  technicality of the way you've, you've worked with data, you've also taken the onus of creating an open platform, sharing back the technical learnings to the community.

And I'm sure there are many more initiatives that you're doing in that regard which we haven't covered yet. So thanks a lot for sharing again. It has been a real pleasure to have both of you on the podcast, Praveen and Sujit.  Thanks a lot for sharing these insights. I know we've only spoken for 30 minutes, but there's been a lot that we've covered.

And. I think there's been a lot for me to learn from, and I'm sure everyone who's listened has also had a lot. So thank you, Praveen. Sujit.