Beyond barriers: How AI is overcoming language barriers and driving inclusion

Podcast Transcript

[00:00:01] Karen: Hello, everyone. I'm Karen Dumville, your host for today's episode. Today, I'm joined by Sachin Malhan, co-founder at Agami, and Vinod S., head of Digital Public Goods and Infrastructure at Thoughtworks India. Together, we'll be exploring the innovative and impactful work behind Jugalbandi, an AI-driven initiative aimed at overcoming language barriers and empowering communities in India. Without further ado, let's dive in. To start us off, can you both introduce yourselves and share a little bit about your roles and how you have worked together? Let's start with you, Sachin.

[00:00:39] Sachin: Yes. I am one of the co-founders of an organization called Agami. Agami is an India-specific organization. Our mission is to bring greater innovation in law and justice. We feel that this is a critical domain, access to justice, legal services, and it hasn't seen as much innovation as it should have. We were set up around seven years back with the objective of building an ecosystem of innovators that could ultimately transform how law and justice operates in India.

Of course, very quickly, we could see that AI would have an absolutely transformative impact on the law and justice space because it's very rules-based, it's a lot of operations and processes. We felt that building an initiative that could create open, usable AI tools for innovators in law and justice would be vital. That's how the OpenNyai mission-- For those who struggle with that name, it's OpenNyai, not OpenAI [chuckles]. There's a little N-Y in the middle of Open and AI. Nyai in Hindi means justice. OpenNyai is open justice, and you could also say OpenNyai for justice.

That's the mission that came out of Agami. That's a mission through which we have collaborated with Thoughtworks from day zero as our technology and data sciences collaborator. We've been building things as a part of that mission from day zero, which was about three years back.

[00:02:15] Karen: Wonderful. Thank you for that. Vinod?

[00:02:18] Vinod: Hi. I've been with Thoughtworks for more than 12 years. At Thoughtworks, I've been responsible for our work in the digital public goods infrastructure space. In that capacity, I had the privilege to work with Sachin and his team at Agami as well. What I do find quite interesting in this space is, it's almost counterintuitive, but we end up working with a lot of real cool tech. Agami is one case in point, but I've got a lot of other cases also where we've been using GenAI, we've been using blockchain as well.

It's actually quite interesting because if you have to bridge the digital divide, you have to actually adopt even more deep tech. That's something that I've come to understand. Yes, it's been great to work with Sachin and his team on Jugalbandi and everything else that Jugalbandi is starting to promote as well.

[00:03:19] Karen: Great. Let's get into some of the details. Sachin, can you explain the concept of Jugalbandi and how it's leveraging AI to address social challenges, particularly in overcoming language barriers?

[00:03:32] Sachin: Jugalbandi is one of the open technologies that have come out of the OpenNyai mission, the Open AI for Justice mission that I was talking about. Jugalbandi is one of the public technologies that has come out of that. One of the open source technology stacks that has come out. The intent is to empower innovators.

Let me explain what this does. in India, one of the biggest problems is access. There are many reasons why people don't access or can't access information, can't access services. What the studies have found is one of the biggest reasons is that the languages of those services and the information is not the language they speak and understand. One other thing, not just is the language, not the language they speak and understand, because India is a melting pot of languages. It has so many different languages that people speak. Almost every few miles, the language changes.

Over and above the language challenge, there's the complexity challenge that the interface of that information or that service is just not clear to them. It's not just a language challenge, it's a complexity challenge. Now, Karen, the issue is that the language and the complexity vitiates the whole effort. Suddenly, almost everything becomes inaccessible to 97%, 98% of people because of language and complexity. It doesn't matter how good the design of the scheme is, the government scheme, doesn't matter how good your service is, doesn't matter how compelling the information is, if it's not in the right language, if it's too complex, person is not getting it. That is a huge disabler.

The idea of Jugalbandi, and Jugalbandi is a word that means something happening in tandem. It comes from Indian classical music, the idea of two musicians operating in tandem. Here, what's operating in tandem is the language AI, the AI that is the language translation, voice-to-voice AI, and the Gen AI, which is taking any piece of information and can ask any question to that piece of information and give an answer like you're a 5-year-old, a 7-year-old, a 10-year-old, a 12-year-old, a 15-year-old.

Think about that combination of the language AI and the Gen AI. The language AI solves the language problem, the Gen AI solves the complexity problem. Now you put that on top of any domain, any use case, and suddenly you've got accessible housing, accessible justice, accessible health, accessible government schemes. That's the thing here. The idea behind Jugalbandi was, could Jugalbandi be a public technology that innovators anywhere inside government, outside government, private industry can slap on to their systems of information and services and suddenly bridge that massive last mile gap?

[00:06:36] Karen: That makes a lot of sense. That sets me up for the next question, which is, what specific civic society initiatives has Jugalbandi been involved in, and how has AI been integrated into these projects?

[00:06:50] Sachin: When we started off, the demonstrative use case that we used was government schemes. India has a massive number of government schemes, hundreds of millions of Indians, I would wager between 600 and 800 million Indians, which is a huge number, like 60%, 70% of the country, access some form of government scheme or the other. It might even be more, Karen, but at least this many access some or depend on some government scheme or the other.

Truth is majority of them struggle to the complete benefits of these schemes or even know about them. One, they don't come to know about them. Two, they cannot participate in them. Both the language challenges and the complexity challenges. We started off by saying, imagine if you're a farmer somewhere speaking your local language and you can talk into your phone and understand how this thing can benefit you or what can benefit you and how it can benefit you.

Maybe it's a scheme to help you build a house, a scheme to protect your property, a scheme to get your daughter educated, could be anything. We began by showing Jugalbandi as an interface for, say, government schemes. Since then, it's that same point I was making earlier, whether it's accessing legal aid, asking your questions around some legal issue that you're facing, your family, your property, could be domestic issue, could be an employment issue, or it's something to do with, say, India has a huge migration challenge, which is people are migrating from rural areas to urban areas for livelihood.

Now, you're not speaking the language of the new place that you're going to. Imagine being able to ask in your own language where can you get your daughter educated in this new city, where can you get affordable housing, how can you get a legal aid support there. The use cases are so many. We as a justice-focused organization are particularly excited about the use cases in the law and justice domain, but, Karen, it doesn't stop there. It's a public technology that solves the access and complexity challenge, which extends way beyond the justice domain.

[00:08:55] Karen: Amazing. There's no end to the amount of applications for this. What were some of the major challenges you encountered during the development and implementation of Jugalbandi, particularly in terms of addressing linguistic diversity and literacy challenges?

[00:09:13] Sachin: I think the first thing is that when we say the language layer. Let's assume, Karen, that that complexity challenge, which is a large language model asking us a set of complex information questions and getting simple answers, that can be solved quite easily because today people are familiar with going to ChatGPT and saying, "Explain to me theory of relativity like I'm a five-year-old," or, "Explain to me Pascal's law," whatever.

I think that everyone can understand that complexity, that you can ask complex information, simple questions and get answers. That is clear. On the language AI side, on the translation side, the major languages, Karen, sure, Hindi, maybe Bengali, maybe Gujarati. Maybe these, there are very good voice-to-voice models, and the government of India has a mission called the Bhashini mission. Bhashini has created open public infrastructure for these languages. It's got all these major languages that they've actually already created the APIs for. They've already created the infrastructure.

I think the challenge is that the major language is always going to be better in their function than the more minor languages because there was more data that has been used to train the AI on the major languages. The minor languages - the minor as in maybe beyond the top five - the quality needs to be continuously worked on. Now, if the quality is not so great, then what happens is the person asking questions can sometimes get frustrated if the quality of the answers that they're getting in their language is not great. I think the major challenge is to continuously improve the quality of the language.

The second challenge I would call out, Karen, is the reality that today, access to LLMs, large language model, is still expensive. The computer is still expensive. What happens is that if you're looking for use cases where millions of people are suddenly going to be using it, then who's going to bear those costs? Nonprofits could struggle to bear those costs. At some level, we have-- I know this is not a problem unique to our domain, but because we are talking about large social impact use cases, this can be an issue. That's the second challenge, the challenge of cost.

The third challenge I'll say is that, look, the challenge of classic entrepreneurship. How is the person at the ground level going to really use the solution? We are imagining them having a simple conversation with their phones, but could they even struggle to have that? Is it that they need human support? Do those solutions at the ground level need to actually empower local stakeholders as opposed to trying to replace them? A local community worker, local social worker, I think we need to start thinking very creatively of hybrid solutions, working with ecosystems or actors. That is the difference.

Rather than point and shoot and say AI is going to solve this problem magically, think of it as an empowerment for existing ecosystems of actors. I think those are three challenges that we need to be mindful of as we try to apply this.

[00:12:26] Karen: Makes a lot of sense. How does Jugalbandi ensure cultural sensitivity and accuracy in its AI-driven solutions, especially when dealing with diverse linguistic and cultural context?

[00:12:40] Sachin: Part of this answer, Karen, lies in what we said about the ecosystem approach. You see, if you're sitting in Delhi or you're sitting in Banglalore and you're saying, "Here's this solution," and the anonymous Indian living in Chhattisgarh or in Bengal is going to be able to use this, and-- I think you are being naive. The way to look at it is you have to work with entrepreneurs who are more sensitive to their contexts.

It is the entrepreneur in Bengal who understands that women there who need domestic violence support, how can I use this for them? How can they talk maybe into their phones using WhatsApp, get simple answers, alert their local social workers to challenges? That context sensitivity is coming from that entrepreneur. This goes back to the beginning that this is public technology, Karen. We cannot be the end innovators. The end innovators are the ones who understand context and can learn rapidly on the ground how this looks. I think it's important to see this as public technology enabling innovators who have use cases in mind rather than thinking of this as a product or a solution to shoot at from a long distance.

[00:13:51] Karen: Yes. That makes sense. How do you ensure equitable access to tools and services, particularly in regions with limited internet connectivity or technological infrastructure?

[00:14:03] Sachin: [chuckles] These are the questions. These are the fundamental questions of applying AI at scale, and I would say it'll demand so much creativity. I don't think we have all the answers, but I can share with you a couple of thoughts that we could deep think more about. One is, can we put these tools on IVR so people can call and interact as opposed to maybe using WhatsApp? Even though the WhatsApp penetration is huge, even then there are people not using smartphones who may need to call and interact with the system, and therefore, using IVR systems, which have been around for a long time, which can be connected at the backend to the technology. One is using IVR.

The other thing is, as I said, empowering local stakeholders as opposed to thinking you're going to get to the end person. Empower the local because they might be able to get to somebody who's more proficient, who's already somewhat of a local changemaker. How do you empower them to use? The third is technological innovations, Karen. Technological innovations like models that one day sit on people's phones, so if they don't have connectivity, Simple technologies, the voice technologies of their language sitting on their phones.

I think those kind of challenge. You have to innovate at all three levels in a country like India. You have to innovate at the community level, you have to innovate at the technology level, and you have to innovate at the infrastructure level. It's not a single-shot thing. What it does require is just avoiding hubris of technology or techno solutionism and trying to see it as an empowerment capability.

[00:15:56] Vinod: Karen, if I may add here. Not so much of the challenge, but maybe audience outside of India may not appreciate this, but the very reason why we adopted WhatsApp as a medium of interaction is because most number of people in India already use WhatsApp. They're comfortable using WhatsApp and comfortable using WhatsApp to send voice messages and listen to voice messages with each other. From an adoption perspective, the thought was this is a plus one in a sense. I'm used to dealing with this medium, and so I'm putting this LLM and Bhashini and put that together into a medium, which is WhatsApp, which I, as a end user I'm comfortable with, and that starts adoption.

Why I'm bringing that is, when Sachin spoke about Jugalbandi about voice and Gen AI, I think there is also this third element of that finance-

[00:16:52] Sachin: Absolutely

[00:16:52] Vinod: -interaction medium. The last mile is what the user is already comfortable with. This is just a plus one as far as their adoption is concerned.

[00:17:03] Sachin: I think I would actually stretch that and say that, Vinod, you pointing out to a third critical aspect of the technology. We spoke about the language AI, we spoke about the Gen AI, but I think Vinod very correctly said plugging it into something that's already being used by people as opposed to having to package it and separately get adoption. I think that's the third critical pillar, which is appropriate technology really, right?

[00:17:28] Vinod: Yes.

[00:17:29] Karen: Clear, and just a lot of complexity on a lot of different levels and a very broad ecosystem. Looking back, in hindsight, what key lessons have you learned from your experiences with Jugalbandi, and how do you plan to apply these lessons to future projects and initiatives?

[00:17:47] Sachin: This is not the first public technology or public resources in AI injustice that we created, but this is definitely the first technology that has really captured public imagination and gone to these very large scale, large impact use case implementations. There's been a lot of learnings, Karen. I think one learning is that when we created this, we assumed that people would take it and run with it, but the capacity to implement these things, that capacity is a huge missing gap because you're not dealing with massively technology-proficient organizations.

It's not that they're technology illiterate, but they're not massively technology proficient. Today everyone's playing a catch-up game. The effort that is required to really connect the dots and support people to implement these capabilities, massively underestimated that effort, Karen. I think that's one learning that we are trying to correct that, and we've created now, in partnership with Microsoft Research, something called Jugalbandi Studio, which allows for you to just implement a bot using Jugalbandi capability in natural language. You can literally write down what you want and it creates the bot for you. Avoiding any need for a technologist to really deploy it for you and those kind of things.

We are trying to solve that, but that's a big learning, that there's a gap there, and we have to pay much more attention to implementation. That's one gap.

I think the second gap is that technologies, the individual pieces are going to go on becoming redundant very fast as the technologies get better and better, so what really is the value? I think the value is in creating a combination of things that is relevant for the context, the combinations of language and WhatsApp and Gen AI combined with some privacy filters, combined with some help on how you design your prompting. All of these little, little things that become use case relevant, become context relevant, that package is where the value's at. The combinatorial innovation is where the value's at, not in the retrieval augmentation system or-- Not in those isolated things alone.

I think it's important to stay focused on that. That's where the real value. The third thing is being better at our ecosystem building. I believe that's a strong point, but we like to say it because we believe that that's where everything is, our ecosystem of innovators and stuff. I'm sure Vinod has more to add there because we've really slogged through this journey together. Vinod, do you want to chip in here?

[00:20:28] Vinod: No, sure. I think one of the learnings, and I'm coming from a technology standpoint, one of the learnings is-- Sachin spoke about privacy data, perhaps hallucinations, et cetera. How do you come up with a framework that can address some of these aspects and put that in a way that the solutions that you deploy would be acceptable for that context without creating a lot of risks? We came up with this framework called the Post RAG framework which we think can be applied in multiple scenarios not just in the public goods space, but also in other corporate spaces as well. Definitely, that has been one huge learning as we started working on this.

Like Sachin said, it's been a journey. We've been on this for at least three years now. We've learned a lot along the way. I think we've been able to find various means to make sure things work. We've been able to quantify some of that learnings. We've been able to create certain patterns, and we believe when we pick up another similar project or another project, even in the Gen AI space, we think we'll be able to bring a lot of those to bear as we start executing those.

[00:22:06] Karen: Sounds like an incredible amount has been achieved in three years. Just switching gears a little bit, we'd love to hear about another significant project you've been involved in, the development of a specialized AI model aimed at assisting with specific legal tasks in India. Can you share a bit more about this project and how it's different from other AI models and how it's been beneficial for legal professionals in their day-to-day activities?

[00:22:35] Sachin: On a lighter, maybe more musical note, I must show you a little connection between Jugalbandi and Aalap. Jugalbandi is when two musicians, specifically in Indian classical music, jam together. Aalap is when a musician improvises on a particular type of music. That's the meaning of Aalap. While Jugalbandi is named Jugalbandi because the language AI is doing Jugalbandi with the Gen AI, Aalap allows the lawyer or the legally-minded paralegal to improve their work. That's why the origin of the name. Aalap stands for AI assistant for legal and paralegal functions, Aalap.

Aalap is a bit of a demonstration, Karen, because you see right now, one of the amazing things about GPT, and especially GPT-4, is that it's really proficient at even legal text. It'll do summaries, it'll extract timelines of data from facts. It'll do a bunch of stuff. It is because it's trained on such a large body of general text, which includes some significant amount of legal text. Then GPT-4 is expensive and it is not custom made for legal functions. I think what the world is discovering is, you can have less expensive models, maybe even open models, that can be fine-tuned for legal function.

That is a huge possibility that that model won't be relevant for something else, but it could be fine-tuned for legal function. Here with Aalap, we wanted to demonstrate that. We wanted to take a model, much smaller model, in this case, Mistral, and I think Vinod can correct me, but it's about, I think, 7 billion parameters, much smaller than GPT and an open model. Can you bring to it a set of legal data, which we did? We collaborated with a bunch of organizations, God bless them, who gave us data and fine-tuned it to show that for certain legal tasks like argument generation, analysis, timeline creation, it could perform decently compared to its start point, compared to where it would be without any tuning.

That was the objective because what are we trying to see, Karen? Not to say that here's something out of the box and you can use it. No, but seeing that if you continue down this path, you can take an open model, a much smaller model, and actually bring data to it and fine-tune it and create a professional aid or a paralegal aid of pretty decent standard. Let's continue this journey. That's the message we're stating with Aalap. When we publish the paper and the data sets and everything, that was the intent behind Aalap to show that something like this is possible.

Now, how much we build on top of that, do we put out another version of this and take it to a high level of accuracy, Karen, that's something we have to strategically think about. That was the intent of this first iteration of Aalap. Vinod, please, there must be technical aspects that you are dying to color in.

[00:25:57] Vinod: Yes, sure. Not so much technical aspects, but Karen, what we definitely are seeing is while we've got generic large language models out there, there are different organizations who are trying to create very domain-specific models as well. For example, Bloomberg is creating something which is very focused on the finance field, et cetera. Like how Sachin alluded to earlier as well, using these models are very expensive. Using a generic model to do everything that you want, it's perhaps more expensive to keep using those models. We think it's very productive, but then if everyone starts using large language models, I don't think it's going to be really good for the world.

It's going to be imperative that we start figuring out models that are much smaller in footprint that can actually reside even in a mobile device, for example, and then start using those. Aalap perhaps is one such end-all in the legal space. Of course, we've used Mistral. If you get to keep working on that, we can see how that can be refined and how that model can become even smaller in terms of footprint so that it becomes far more effective for us to start using this.

What we see is across the board, many organizations, some of our customers are also starting to refine models, fine-tune models with the idea of making them smaller and more efficient so that their compute, their power, the electricity, all of that starts getting much more affordable for what is what is required.

[00:27:34] Sachin: You see, when you are looking at some sensitive use cases, like for instance, certain kind of legal system deployment, you also want to consider a model that you could potentially put inside your organization, not necessarily use an API to a gen real model. That's another dimension here, that what is possible to also literally implement inside your own organization. That's where the open models fine-tune to your data or domain relevant data becomes so important.

[00:28:05] Karen: Very good. Talking beyond government programs and schemes, in what other areas do you see potential applications for Jugalbandi and Aalaps, AI-driven solutions both within India and globally?

[00:28:20] Sachin: See, with the Jugalbandi, practically, anybody who has a language and complexity challenge to solve and who wants to do it using open technology, that's a very large cut. Think about it like, I was recently in Spain, and a large chunk of Spain speaks Euskara, the language of the Basque region. There are so many migrants in Europe who don't speak particular language. Imagine in Germany, there are a whole bunch of migrants who don't speak German.

I think this language and complexity challenge is so profound that a technology like Jugalbandi, and hopefully many more, Karen, not just Jugalbandi, because we are a public mission. For us, it's more important that possibilities become realized rather than only Jugalbandi gets used. Hopefully, people take it, improve it. That's what combinatory intelligence is. To Jugalbandi, there's really no limit in that sense because that's the significance of what is being attempted here. Correct?

I'm also really keen to see it trigger the creation of custom language models for smaller languages. Imagine a tribal language. Imagine a language of aborigin communities in Australia are able to actually create a language model and suddenly you have so much more pride and intimacy with whatever you're interacting with. This can unleash something-- Think about it, few years back, we thought the world was moving towards greater uniformity. You got to learn the same languages to really make it. What if we could flip it? What if we say that the world is actually moving towards radical diversity? That I think is really exciting.

I think on Aalap, we think that if tools like Aalap could empower local lawyers so much more, empower paralegals so much more, maybe make citizens feel like paralegals. Just that overall empowerment is a form of decolonization, Karen, because at the end of the day, all the intermediaries in systems like lawyers and accountants and all, they pride themselves in speaking a certain complicated code of words or language. If you can empower people, I think it's good for everybody. It's good for the lawyers, it's good for the citizens. I think seeing more tools like that come out into the domain is going to be really exciting.

This applies to any common law country, Australia, the UK, Canada, because Aalap will be finetuned on common law, the Indian legal documents, but essentially with this strong common law connection. I think just showing how Aalap is done and what it's capable of allows many other innovators to build on top and say, "Hey, we should do this." We're excited about its triggering effect as a catalyst to other public technology creators. In fact, we'd love to see more people who say, "I'd like to create a custom model in law. What can I learn from the approach you took?" That's how we're thinking about it.

The last thing I'll say is our collaborators on the mission, organizations like Thoughtworks, the learnings they've had on how to go about doing this, that itself being a tremendous leap, where they then take those learnings out to other public-minded operators.

[00:31:37] Karen: Indeed. I'm sure Vinod can probably speak more to that.

[00:31:42] Vinod: Yes. Karen, not many know, but we've taken Jugalbandi. We have tried to create a version of that called Jignyasa within Thoughtworks. It's still in the works in terms of getting deployed, but I already have some pilots running. The idea is very similar to talk to your documents. If you are working on certain RFPs, if they're very large documents that we need to query and then get answers for, et cetera. Thoughtworks is a corporate organization. You got one real use case right there. We're trying to see how we can adopt Jugalbandi's open-source stack and then use that for our internal purposes. That's number one.

we are very closely working with another large financial organization where we are setting up something similar within their internal setup, where you could use similar stats for their product engineers and other folks to query the system, given a lot of information and data, and then get precise information. We're also taking that forward to see how, if you were to share code into that system, how that can come back and start giving us responses in terms of what's wrong with the code, what other unit tests or automated tests can be added to it, and so on and so forth. The opportunities are immense. That is number one.

Number two is user interaction. I think we are setting the stage for voice-first mediums of interaction. I think there is a large demography out there who would really take corporate services. If you start interacting with them, just basis Voice. I know of several organizations who are taking that approach where I query, I need to make a payment, I need to make a search. I'm not using keyboards at all. I just use my voice and my site to be able to do all of that. I think that's going to create a plethora of new organizations, new ways of reaching out to customers, and hence, new opportunities for corporates as well.

[00:34:01] Karen: No end to the number of applications. I'm sure your minds are just constantly worrying with ideas. Additionally, how do you foresee tech advancements such as integrating large language models, shaping the future of conversational AI platforms?

[00:34:19] Sachin: The work we've done is only a drop in the ocean, hopefully, a big drop, but a drop in the ocean of the work that's being done on conversational AI and language and interfaces. For instance, there's also amazing work on Access AI that is transforming how you access and interact with the system being done by Sarvam, which is a startup that the former co-founder of OpenNyai, as well as the lead scientist and also co-founder of OpenNyai, have created about six months back.

They are a startup that has gone and already shown the promise of how interfaces, Karen, will be transformed thanks to AI. I think that we are going to see so much change in how you engage with the Internet, how you engage with information and services. I think it's just the beginning, to be honest. 

Going forward, we're going to see huge jumps in the access part of this. I think next, not next two, three years, next six months, next nine months, we're going to see these innovations drop and transform our understanding of what it means to access something.

[00:36:13] Karen: It's moving so quickly. What has been the role of partnerships in your success with Jugalbandi?

[00:36:20] Sachin: Jugalbandi is entirely the consequence of co-creation. Not just collaboration, but co-creation. I would say right from the get-go, we knew that no one organization had the capacity to build something like this. How do you create the connective tissue for multiple organizations? In this case, the founding collaborators of the OpenNyai Mission were Agami, the organization that I represent because of our commitment to innovation in law and justice. Ake Step, a leading nonprofit focused on innovation in education, Thoughtworks, and of course, the leading law school in the country, the National law School of India University in Bangalow. Those are the co-collaborators.

Really, everyone worked incredibly close together to make this happen and co-create itin the sense of reimagining the vision, building things together, sharing knowledge and resources, and really being one fluid team of teams.

Subsequently, many other collaborators, like Microsoft Research, has been an absolutely critical collaborator that has seen the potential of Jugalbandi and then said, Let's make it really easy for people through something like Jugalbandi Studio implement these technologies at the touch of a button or in natural language using their own frameworks. I think we've just seen it. Bhashini, the language mission, without which, how would the translations happen?

I think the degree of, and there are many others, to be honest, the innovators themselves who've actually implemented the use cases, some of them particularly close to our hearts because some of the earliest implementations like Bandhu for migrant workers, and also, Civis in the public consultations space, many, many, many, Karen. The short answer to your question is, the core innovation inside OpenNyai is actually not the technology. The core innovation is what does it mean to create a co-creative mission. What does it mean for multiple people to hold a vision in their own distinct ways and do their own special things at the same time achieve that vision together?

[00:38:28] Vinod: I would want to second that. Karen, I do want to highlight, in Thoughtworks, when we say we want to do a project, we say we want to bring together a cross-functional team. A partnership is just an elevation of that. You've got partners who bring different skills but then combine with a common purpose. The second point I would want to end us what Sachin said. I think our mission was very much there even before Gen AI became a fad. It's just that with Gen AI, we've been able to progress things much faster. We've been able to pull these things together. I think that common purpose and mission has been very crucial for us to be able to take this forward.

[00:39:09] Sachin: If I can just add here, the work of the early collaborators and some of the early data scientists from Thoughtworks was absolutely instrumental. I'm not one for necessarily calling out individual people, but they know who they are. Really, it's been many people who've stepped up and really been true cofounders of the mission.

[00:39:34] Karen: Thank you, Sachin. Just wrapping things up here. Looking ahead, what are your aspirations for the future of initiatives like Jugalbandi and Aalap? How do you envisage them continuing to drive positive social change through AI-powered innovation?

[00:39:52] Sachin: I think that goes to the heart of our mission. Really, what does it mean to have a mission that stands for OpenNyai in justice? It means that there's another way of working. There's another way of working where we can collaborate, we can keep things open, we can share our resources, we can build powerful technologies that enable each other. Maybe we don't have everything in our hands. Maybe there's still going to be people who are going to be creating the best LLMs. There's a whole bunch of other stuff that can be kept open, can be shared.

If we can do this with innovators, as opposed to doing it in some walled garden somewhere, then we make ourselves domain-relevant. We ensure that those are real needs that we are addressing. Real issues of safety can be addressed. You can't do that inside a walled garden, but on the ground, you know what the safety issue is, and that feedback travels right up to the collective.

Really, Karen, it's about ensuring that we continue to operate as a collective. We have those grassroots understanding and grasstop understandings, continue to create the critical public technologies, and really build new culture in this space. I think this is what it's really about. It's really about a form of culture. You know you can only see the importance of that when you compare it with cultures that are monochromatic, where somebody is rolling out a product somewhere with the top AI, nobody knows what it is, and it's supposed to do great things. On the other hand, when you have a culture, lots of people win, lots of needs are addressed, lots of feedback is taken, and I think that's what we want to see.

[00:41:28] Karen: Thank you. Vinod, is there anything you'd like to add to wrap this up?

[00:41:33] Vinod: I, again, completely endorse what Sachin said. I may be repeating this, but ultimately, it comes down to adoption. You've got this platform you want people to adopt, you want people to find use of it, and you definitely do want to see if this goes not outside the borders of India as well. The good news is we've had several inquiries. We had some interest from the World Bank where we had demonstrated. Some countries have also come and seen this. Our aspirations are that it gets adopted. People find it useful and it makes a real change to people's lives.

Let me also say it's been a privilege to work with Agami. We are a technology company, but we do have a heart. It's great to work with a company like Agami to make some of this happen. At the same time, being a technology company to be able to hone our skills on some of the latest technologies is just the icing on the cake. Thanks a lot, Sachin, for this opportunity.

[00:42:36] Sachin: Thank you.

[00:42:38] Karen: Thank you both so much, Sachin and Vinod, for your time today. It's been truly interesting and inspirational to hear your story, and you both should be incredibly proud of what you're working on. Thank you so much.

[00:42:53] Vinod: Thank you, Karen.

[00:42:54] Karen: Thank you.

[music]

[00:43:03] [END OF AUDIO]