Like at that point, it was one of the most expensive things Indian households have, right? It's more expensive than the TV and the fridge combined. So, I thought this is done.I'm never getting out of it. But thankfully, nothing major had happened. Things worked out fine.And after that, I think I developed this love for disassembling things, building things from scratch. And yeah, that started my journey into learning more about hardware and software and so on and so forth.

Host: Yeah, well, that's truly interesting, Rabi. Moving ahead to our next one. So, you have been at multiple businesses now. You get bought out by Google as well. How did you first get involved with AI? And what inspired you to launch Thesys?

Rabi: Yeah, actually, the story goes like my last company, when the whole ChatGPT movement happened. We decided internally to shift our strategy. We were doing probabilistic machine learning for the longest time.

And we decided to switch our strategy to more MLM-based. And since it was like a major rehaul of almost all things, I also got pulled in. And yeah, the next six to nine months, we spent a lot of time working with open AI models, trying to build a customer support chatbot, etc.

And I realized that there are still a lot of gaps in how great technology is. But in terms of building a successful product, it still needs a lot of good engineering. And that's how I got involved with AI.

That's how I realized there is a huge opportunity here to build something meaningful, build something rounds up and get AI adoption more across the enterprise use cases. And that's how I got started with Thesys. And that's why I started Thesys.

Host: Yeah, AI has a great potential right now. And there are multiple spectrums of AI that has been unexplored, and you are keeping along. So, it's truly great.

Rabi: Yeah, absolutely. There are still lots of gaps in terms of taking the technology and producing it into a product. Thesys aims to fill one of them.

But I think over the next few years, we'll see more and more of these applications that are actually truly powered by AI. And AI is not just an add-on, just a bolt-on on top of your existing apps.

Host: So, let's talk about some tech now. In your opinion, why Gen UI? People are talking about computer vision, enhancing LLMs. You see video generation with Sora. All these talk about AI-powered robotics. Why did you decide on Generative UI specifically?

Rabi: Yeah, so honestly, people forget that LLMs are not the product in itself, right? LLMs are finally just a technology, and we still need to build a meaningful product out of it. So, I know it doesn't sound as sexy as saying that, oh, we are increasing the token count, or we are reducing the cost of tokens.

But at the end of the day, your users don't care about the LLM. What they finally care is about the product, how useful it is, intuitive it is, right? And GenUI is a big part of that problem.

Text is great, but text is not always the preferred way of communicating with software. Like over the years, all the innovations, all the breakthroughs that have happened in designing better interfaces, they have kind of got sidelined with AI in general. And that's what we are trying to fix.

It's not as exciting as, for example, working on a new model or working on a new thing, but it is absolutely necessary for you to be able to ship these interfaces in front of actual customers.

Host: Yeah, certainly Generative UI is truly fascinating and undiscovered layer of AI, I must say. So, going forward, Ravi, how do you see the Generative UI conversation evolving? Are developers asking for this? And what are the core strategic problems that you are solving right now?

Rabi: Sure, so in terms of developers, like it obviously stems from even my pain point of building AI software in my past companies. And right now, developers are kind of tired of just doing on some sort of an interface over their text-based responses, right? I mean, a lot of time and energy has been spent in trying to perfect the backend, but the frontend has been gnarly neglected and there aren't a lot of good ways for you to build an actual meaningful interface powered by AI.

So that's in terms of what developers are asking for, like a better way, a more intuitive way, a more flexible way of building AI software. But at the same time, the strategic importance of why you should care about Gen UI is mostly like the iteration speed, like the time it takes to ship software. It is like the promise of LLM says it will go down and down more.

And at some point it should feel like it's almost like writing a story to ship software, but without a usable interface, without a meaningful way to interact with that software for your customers, it's not going to happen, right? This is why generative UI is important because it not only gives the developers a more flexible way to ship software, but it also reduces your time to market, your increases or iterations speed in shaping these interfaces.

Host: Actually, you take on this quite fascinating. Next up, I must congratulate you on your recent launch of C1 by Thesys, the generative UI API. Can you share some more insights about the aim behind building C1 by Thesys? It's core functionalities and how it's set to transform the way AI powered products are built and experienced.

Rabi: Yeah, yeah, absolutely. So, the best way to think about C1 would be like what Stripe did for payments or what Twilio did for telephoning and communication. Gen UI does C1 by Thesys does the same thing for the interface layer, right?

We help you ship these interfaces live in front of your customers, not just as a static power thing, but something that adapts to your customer needs depending on what they ask, how they interact with it and how you use it is pretty simple.

You use it in the same way you use any other LLM API, but now the LLM API instead of churning out text can churn out live usable interfaces on the fly and just adding a layer of C1 by Thesys on top of it existing workflows, it can be a customer support bot, it can be an e-commerce shopping experience, increases the usability and increases your time to market so much that it ends up increasing your retention and it ends up increasing your engagement with these AI powered interfaces.

Host: That's an amazing milestone for Thesys and it's going great, I suppose.

Rabi: Yeah, absolutely.

Host: Let's approach further. What were the major products or technical challenges you faced while building C1 by Thesys and how did you overcome those?

Rabi: So, the biggest challenge was even defining what this product should look like. I mean, this idea is not new, but there are no practical implementations of doing genuine. We had to go back to the whiteboard and Parikshit and I, and Parikshit is my , he is a designer.

Alongside, we had this thing that we have to build something that we love and we ourselves would have used in the first place if we had it. So just defining the product requirements, the shape of the product, whether it should be an API, whether it should be an SDK, whether it should be code or something else, that was I think the toughest part. But obviously, once you are over that, you still need to work with LLMs and LLMs are not perfect, right?

You still need a lot of good engineering to make sure LLMs respond in a meaningful way, right? It does not break it, like we are talking about generating your UI on the flight, right? So, it cannot break.

Your customers still need to see some valid outputs. You are going to put it ideally inside your own product. So, the generation should look and feel like your own product and should not just look on as another glue code or another bolt-on that you put on top of your product.

All this combined, the last six months has been a very interesting journey, trying to figure out the product shape first, then trying to figure out the roadmap, then working with our design partners and making sure that the form factor makes sense, so on and so forth.

Host: Yeah, it included various stages, I suppose, and the challenges are truly a part of the emerging success.

Rabi: Absolutely, absolutely.

Host: On to the next, what's your take on data privacy, particularly when generative AI generates UI from user inputs, does Thesys ensure data remains safe according to the industry standards?

Rabi: Yeah, I mean, I think that was a no-brainer part of defining the product experience that your customer data or your data is, end of the day, your data. We don't save that data while you're using our API. We don't train on that data.

You have complete control over that data. Obviously, the data is encrypted at transit and everything, but we have made sure that there is no way for us to intercept that data and try to train or do stuff on top of that.

Host: It's truly amazing to know how Thesys is prioritizing data privacy.

Rabi: Yeah, I mean, that was a table stake for us. Again, going back to the analogy of something that Parikshit and I would have used in our day-to-day setup, if I had to use a product like this, I would definitely always make sure that my customer's data is safe, my customer's data is completely in my control, and that's what we set out to build.

Host: Yeah, moving ahead. Thesys is not storing the data, but the API is just processing on top of it. So it really is up to the LLM.

Do you work with all the LLMs? Are you OpenAI compatible in this regard?

Rabi: Yeah, I mean, we are working on extending our support right now. For example, it's built on top of the Anthropic Family of Models, the Claude 3.5 specifically, and we are working on adding more support. But before that, we wanted to actually take a step back and let our users decide let our customers decide which one works best for them.

That's why what we did was we created our own benchmarks. We're going to publish the results with a live benchmark playground very soon. But once that happens, you can see the quality of output difference between different models, and it will definitely work with OpenAI, but that goes a little lower on the benchmarks that we have.

But if it works better in your use cases, and there are certainly a lot of use cases where the OpenAI Family of Models would work better for you, then we can definitely support multiple models and OpenAI is the top of the list, right?

Host: Absolutely. Interesting to hear that. Let's get on to the next one.

What are example use cases people are building with? Do you have customers today that you can discuss?

Rabi: Yeah, I mean, without naming names, I can tell you two specific use cases. One, they're building a customer support kind of board, but instead of, you know, instead of just dumping text and asking the users to type in a wall of text, now they are actually generating refund forms on the flight. They're actually guiding you through the return process.

They are populating fields in your CRM. So that is that we see as one of our bigger use cases, customer success, customer support agent kind of things. And second, there is this multi-billion-dollar public materials company that has built an internal version of the procurement software.

So they have these warehouses all over the world, and these warehouses need to order inventory and they also need an analytics dashboard to see what ordered, etc, etc. So they reimagine this entire thing with AI and C1 was very integral in like helping them ship that in the piece.

Host: That's indeed impressive and surely the key indicators of your success. Moving to our last one, where do you see the conversation extending to AI and the future of tech? What comes next according to you?

Rabi: I think a lot of things in a lot of places, like obviously all the video stuff, all the image generation stuff, they are just going to get better and better. A lot of workflows that were very human dependent are going to become more democratized, like for example, generating marketing content, etc.

But speaking within the narrow limits of how we see software interfaces changing and what that would mean is like the last couple of years, three years have been mostly around improving the backend and talking about in terms of like, hey, we have increased your efficacy rate or we have reduced hallucinations or we have reduced the token counts and stuff like that.

But going forward, we'll see more and more of these AI native interfaces, truly built, thinking deeply about AI as an integral part of your workflow and not just an add-on, not just a co-pilot on top of your existing experiences. And that would also mean a lot more dynamic interfaces, a lot more intent-driven interfaces. And yeah, the next five years are going to be super interesting in terms of software interfaces and how people interact with software interfaces.

Host: Yeah, crucial insights on the future of tech. And that's a wrap. Thanks for being here with us today, Rabi. Your insights into the into and around generative UI and the launch of Thesys and the future of tech are noteworthy. Thank you for being here.

Rabi: Thanks, Rittika. Thanks for having me.

Host: Also, a big thanks to everyone for tuning in to ExtraMile by WisdomPlexus. I'm your host, Rittika, signing off for the day. Stay tuned for more exclusive interviews with the industry pioneers, sharing their thoughtful wisdom around the latest tech and trends.

Until then, stay tuned!