Beyond the paper: How AIResQ ClimSols is protecting urban India from flooding

Startup Genesis & Operations

Let’s start from the beginning. How did the idea of the startup come about, and how did IITGN support you?

Prof Udit Bhatia: The idea emerged as our lab, the Machine Intelligence and Resilience (MIR) Lab, started maturing. As our publication portfolio was progressing and gaining attention, we realised that this work was deeply connected to real-world problems that urban and water infrastructure systems face, and just documenting that in research articles may not take the knowledge to the last mile.

That’s where the idea spawned: that we should have a lab-grown entity, a startup, that could commercialise these solutions in an end-to-end manner. For us it was clear, the research lab brings the deep research component, and the startup takes care of it as a deep technology and scaling it up.

I also drew a lot from my PhD experience at Northeastern University, Boston; that lab has been at the forefront of translational research, and multiple startups have come out of it. So I already had a good sense of how to get something like this started.

I took at least four to five years after joining IITGN in 2019 just to set up my base here. And then I realised that the kind of support, or freedom, I would say, more than the support per se, that an institute like IIT Gandhinagar offers is quite unique. The IIT Gandhinagar Innovation and Entrepreneurship Center (IIEC), Research Park and their folks, they guide you honestly, without giving a very flowery picture, like “Do this and you will be able to immediately raise crores.” That kind of clarity is what I really think matters the most when you are trying to bridge the gap from our academic block to the research park block, because that translation is fairly non-linear.

Dr Divya Upadhyay: I would like to add that the transition from lab to startup became necessary when we saw that cities did not just need analysis reports; they needed usable solutions. In research, a model can end at a paper or a prototype. But in practice, cities need something that can ingest data, run reliably, generate interpretable outputs, and support decisions during high-pressure situations. 

You already had a well-established lab and students. What was the tipping point, the moment you felt the lab alone wasn’t enough?

Prof Udit Bhatia: The tipping point came when the city of Surat approached us to do a flood resilience assessment for their upcoming riverfront project. We took it on as a small consulting-cum-research exercise. But while working on it, we realised: there are hundreds of cities in India and in the global south struggling with similar problems, and that’s where scalable science can help.

The issue is that researchers, especially PhD students, are always chasing novelty. Nobody wants to repeat what’s already been done. But if I wanted to scale the same solution to 500 cities through the lab, I’d be asking researchers to do something repetitive, and that’s a fundamental mismatch. A startup, on the other hand, can focus exclusively on scaling without the burden of publishing novel research papers every time.

The timing also worked out. My own PhD students were nearing graduation and were aligned with this vision. Divya, who is sitting here, was my first PhD student, and she’s now the Co-founder and CTO. 

Dr Divya Upadhyay: For me, the tipping point was when I realised that many of the scientific questions we were solving were already meaningful enough to create real impact outside academia. When I started seeing how cities were struggling with recurring flood problems during my PhD journey, I realised that our work should not stop at publications, and that we need to translate our research into solutions that are actionable, scalable, and usable for the people who actually need it.

You already had a well-established lab and students. What was the tipping point, the moment you felt the lab alone wasn’t enough?

Prof Udit Bhatia: The tipping point came when the city of Surat approached us to do a flood resilience assessment for their upcoming riverfront project. We took it on as a small consulting-cum-research exercise. But while working on it, we realised: there are hundreds of cities in India and in the global south struggling with similar problems, and that’s where scalable science can help.

The issue is that researchers, especially PhD students, are always chasing novelty. Nobody wants to repeat what’s already been done. But if I wanted to scale the same solution to 500 cities through the lab, I’d be asking researchers to do something repetitive, and that’s a fundamental mismatch. A startup, on the other hand, can focus exclusively on scaling without the burden of publishing novel research papers every time.

The timing also worked out. My own PhD students were nearing graduation and were aligned with this vision. Divya, who is sitting here, was my first PhD student, and she’s now the Co-founder and CTO. 

Dr Divya Upadhyay: For me, the tipping point was when I realised that many of the scientific questions we were solving were already meaningful enough to create real impact outside academia. When I started seeing how cities were struggling with recurring flood problems during my PhD journey, I realised that our work should not stop at publications, and that we need to translate our research into solutions that are actionable, scalable, and usable for the people who actually need it.

You registered in November 2024. It’s still early. Has there been a moment where you felt things might not work out?

Prof Udit Bhatia: Honestly, it happens on a weekly basis.

There’s a fundamental difference in expectations between research and deployment. In research, we can acknowledge model limitations, report negative results, and still advance the field. In deployment, stakeholders want answers, with minimal errors. That tension creates what I’d call deadlocks, more often than you’d imagine.

But because the lab and the startup coexist and communicate, we converge toward solutions much faster than if we were working in isolation. It also helps me partition problems clearly: if it’s a fundamental research challenge, it goes to the lab; if it’s a scale-up or deployment problem, it goes to the startup. And technology transfer between the two is facilitated by structured agreements, MOUs with IIT Gandhinagar that clearly define resource usage, royalty sharing, revenue sharing, and compensation. When things are written down unambiguously, daily operations become much smoother.

Dr Divya Upadhyay: In research, you can spend months refining a model. In deployment, stakeholders need timely and dependable outputs, even while the system is still evolving. One of the hardest parts is deciding what is sufficiently robust for real-world use, while still pushing the technology forward. But that is also what makes this journey meaningful to provide solutions under real constraints, not ideal ones.

Projects & Client Pipeline

What was your first pilot project as a startup?

Prof Udit Bhatia: Gurugram. It started as a pilot, but it’s now a major project. Gurugram has a severe urban flooding problem for years now given its saucer shaped topology, extremely dense urban structure, and economic boom that fuelled it further. We are now working with Gurugram Municipal Corporation under a multipartite agreement involving the MIR Lab, the startup, Airawat Research Foundation (IIT Kanpur) and the corporation. It’s become our primary use case and will eventually become a core product. Now, we are on the verge of securing another major project from a capital city. 

How did Surat first approach you? How does that pipeline of government clients typically work?

Prof Udit Bhatia: Government agencies (municipal corporations, water departments, dam safety organisations) regularly approach IITs for specialised technical consulting. Over time, our lab has built a very tightly knit community in the water space. We’ve become a sort of technical epicenter.

Our doors are open even when there’s no formal project. Officers walk in and say, “This is what we’re facing right now, what do you think?” Those conversations became the nucleus of our research problems; and eventually, they revealed a huge gap in the market. We realised this wasn’t just about serving these agencies as clients. The question became: how do we scale this to 500 cities, multiple states, and eventually countries with similar challenges? That shift in thinking is what shaped the startup’s mission.

Dr Divya Upadhyay: During these conversations, what we have learned is that government stakeholders are not looking for technology for its own sake. They want systems that can help them to make real decisions by knowing in advance where flooding may occur, which roads may become inaccessible, what infrastructure is vulnerable, and what action needs to be taken. So from our side, the challenge is not just developing sophisticated models, but packaging the outputs in a way that is decision-relevant, interpretable, and operationally useful.

Core Technology: Physics-Informed AI

Why physics-informed AI specifically? Why not pure ML or traditional hydrological simulations?

Prof Udit Bhatia: Both approaches have real limitations at scale.

Pure machine learning requires large, consistent datasets. But when you’re working across 560-plus cities, you won’t get data at the same granularity or resolution everywhere. Missing data is the norm, not the exception. Most ML approaches need large volumes of data to identify patterns; so that breaks down.

Physics-based models, on the other hand, are highly interpretable and built around well-understood principles, such as conservation of mass, momentum, and so on. But these models also have a heavily parameterised component, where we’re essentially fitting equations that we don’t fully understand to real-world behaviour. That’s where they break: they don’t generalise well across contexts, and they’re computationally expensive. Running them in real time for multiple cities just isn’t practical.

Physics-informed AI gives us the best of both. Physics provides structure and robustness where data is limited. Machine learning fills the gaps in speed and scalability. The real innovation lies in knowing when to rely on physics and when to bring in ML; that dynamic balancing is our core technical advantage, and it’s what allows us to deliver hyperlocal, real-time flood intelligence.

The cities which are part of your project sphere are distinctly unique: Surat is coastal, Gurugram is inland, Kozhikode has river confluences and landslide risk. How do you adapt?

Prof Udit Bhatia: Context matters enormously. I cannot apply the same solution to Gurugram that I applied to Surat, and I cannot apply what I did for Gurugram to Western Railways either.

What physics-informed AI allows us to do is identify common patterns across cities while resolving the specificities of each one, without sacrificing speed or accuracy. We also don’t hesitate to do what we call micro-zonation, which is breaking a city down into zones with distinct characteristics and treating each accordingly.

The broader principle is this: research methods may generalise, but solutions must be specific. And enabling that specificity requires fast, reliable iteration, which is exactly what our scientific engine is built around.

Speed and accuracy are typically a trade-off in computation. You’re working in a high-stakes domain. How do you manage both?

Prof Udit Bhatia: The trade-off is real in a general sense, but it isn’t a fixed law. Look at large language models; they started slow and inaccurate, and now they’re fast and reliable. The same trajectory is possible when you design thoughtfully.

In our case, physics constraints prevent the model from going outside physically plausible bounds even when iterating quickly, which naturally preserves accuracy. And when you think of the full pipeline (from city modelling to hydrology to water flow), if each layer is carefully designed with domain knowledge and clear quantification of uncertainty at every stage, the whole stack can talk to each other efficiently. That’s where the trade-off breaks down. It’s not universal; it’s specific to how well the problem is architected.

Business Model & Impact Measurement

Most of your clients are government bodies. Their procurement cycles are notoriously slow. How do you stay financially stable?

Prof Udit Bhatia: There’s no single answer. What we try to do is lead with proofs-of-concept, demonstrating real value before any formal procurement. And we try to position our work clearly as high-risk, high-reward engagements, not generic procurement. When extraordinary expertise or technology is involved, most government systems do have enough flexibility to find pathways.

We’re a commercial entity, and we care about revenue and profit, but we’re not thinking like profit hoarders. We’re thinking about sustainable, impact-driven growth. The goal is to keep improving our models, keep scaling, and measure success by lives impacted, not just by the number on a balance sheet.

How do you define success when your models will never be 100% accurate?

Prof Udit Bhatia: We’re not chasing perfection. Accuracies may never hit 1, but they should reach what I’d call the operationally acceptable level. Our real measure of success is: given the data available, what’s the best decision that can be made? And how does that compare to what was possible before?

The benefit-to-cost lens is important here. If a city was previously losing X dollars in flood damage, and our system helps them save a meaningful portion of that, that’s success. We evaluate ourselves on improvement, not on absolute accuracy.

Dr Divya Upadhyay: If our system helps a city identify high-risk locations earlier, prioritise interventions better, or reduce response delays, then it is creating value. In domains like flooding, the question is not whether uncertainty exists or not, it always does. The question is whether we are helping stakeholders act better despite that uncertainty.

How does AIResQ ClimSols handle the responsibility when a prediction turns out to be wrong?

Prof Udit Bhatia: Through transparency, clear agreements, and honest communication of limitations.

We are not guarantors. Cities and floods are stochastic systems with multiple stakeholders and decision layers; no one can eliminate that risk, the same way insurance doesn’t eliminate risk, it only reduces exposure. That’s exactly what we tell our clients.

And we never oversell. If our accuracy is 75%, we communicate 75%, with full pros and cons. If it’s 88%, we say 88%. That courage to be honest about limitations comes directly from a research culture. In research papers, you don’t inflate your results. We apply the same standard in deployment.

If anything, we’d rather disclose higher levels of risk than understate them, because overselling in our domain doesn’t just hurt a business, it under-prepares communities for real flooding events. We’d rather people be cautious and prepared than falsely reassured.

Scaling, Vision, and Advice

Tell us about the ARC Centre; you already had a lab and a startup. What was the need for something like this?

Prof Udit Bhatia: Everything we’re doing is about scaling, not just technically, but in terms of communication and perception. I can write research papers and publish them in top journals, which is very important to test the rigour and reproducibility of models and underlying science, and is crucial for us. But until stakeholders actually see these models in action, at the scale relevant to them, it doesn’t translate into decisions.

The ARC Centre was built to be that space, resembling the command centres that cities and states already operate, but fuelled by real-time science. It’s where the water community, city planners, researchers, and policymakers can come together and visualise models live. It improves both how research is done and how it’s understood.

Honestly, the instinct for it came from how I teach. Even in mathematically intensive courses, I always teach in story mode: rich visuals, clear narrative, no compromise on rigour. Over seven years at IITGN, I noticed that this approach doesn’t just make classes more engaging, it makes students think ahead in the same solution space. Many of my best students chose to stay on for advanced degrees because of it. ARC Centre is an extension of that same philosophy, at a much larger scale.

And it’s catching on; people have started asking us to help them replicate it in their own offices. We’re now advising cities and clients on setting up similar systems and building pipelines so that our models directly fuel those centres.

And ARC Centre 2.0; what’s the vision there?

Prof Udit Bhatia: The next version is about immersive interaction. Right now, you see cinematic screens and live model outputs, and we’ve recently introduced smart glasses for a walk-through experience. But we want to go much further.

Imagine a SimCity-like setup, but it’s your actual city. You’re standing in a 2,000 square foot space, and you can interact with the urban flood problem, the water supply problem, and the drainage infrastructure, all in real time, with actual data. That’s what ARC Centre 2.0 is working toward: augmented reality coupled with our physics-informed models, letting decision-makers feel the consequences of urban choices before they make them. The impact on reasoning and preparedness would be immense.

You were selected under the Government of India’s AI Center of Excellence (AICOE) “Make AI in India” initiative. How significant was that?

Prof Udit Bhatia: It was genuinely significant, not just as a badge, but because of what it represents. This was a competitive process where almost all IITs participated. After a six-month proof-of-concept phase, IIT Gandhinagar’s MIR Lab was selected for the urban flooding component of what is a ₹300 crore national initiative, anchored at IIT Kanpur’s Atal Research Foundation.

One of the mandates of AICOE Sustainable Cities is clear: not only produce research papers but also scale the research to multiple cities. That mandate coincided almost perfectly with the formation of AIResQ ClimSols. It was one of the stars that had to align, and it did.

As for other recognitions (getting these or missing them), we appreciate them, but we don’t read too much into any of it. As researchers, we’ve developed thick skin. Papers get accepted, papers get rejected. Events go well, events don’t. Our mission is bigger than any of those outcomes.

In just over a year, you’ve filed a patent, published in Nature Cities, launched a command centre, and deployed pilots across multiple cities. How do you manage research and startup execution simultaneously?

Prof Udit Bhatia: It’s entirely about the team you build around yourself. I only have 24 hours in a day, and so does everyone else. What makes the difference is the environment you create: one where people can experiment, fail, iterate, and stay motivated through real purpose.

The urgency helps. Every delay brings us one step closer to the monsoon season, and we remind ourselves of that constantly. When you know that Gurugram is waiting for sensors to be deployed, and that knowing which manhole or which area will flood could protect thousands of people you’ll never personally meet, that drives you.

Papers and patents are in the rear-view mirror. When we look forward, the same scale of challenge is still standing in front of us. We measure our success by how many lives were impacted by what translated from the lab to the startup to the field.

What would you say to a faculty member or a recent PhD graduate who wants to start a deep-tech company but is on the fence?

Prof Udit Bhatia: Try, tinker, and dare to fail.

Deep-tech is not like building an app or a service startup where you have clear role models and playbooks. You are simultaneously creating the market, defining a paradigm, and building the product. It’s a long road, longer than most people expect, and you will face more failures than successes early on. The key is to fail gracefully and not give up.

The second thing I’d say is: find your actual USP. It may not be obvious on day one. But over time, ask yourself: what is deep enough and translatable enough that it would create a real moat? That’s the question worth obsessing over.

And when your research actually starts coming into action in the real world, the satisfaction is immense. Nothing quite compares to it.

Dr Divya Upadhyay: I have been part of this journey from the lab to the startup, and what continues to drive me is the belief that strong science should not remain confined to papers or prototypes. As CTO, I think constantly about how to make our research usable in the real world and reliable enough for deployment, flexible enough for different cities, and meaningful enough to support better decisions on the ground. For me, that is the real purpose of technology transfer: not just moving knowledge out of the lab, but ensuring it creates tangible societal value.

AIResQ ClimSols Pvt. Ltd. is incubated at IIT Gandhinagar. The company is currently deploying urban flood intelligence systems in Gurugram and is scaling its Rain2Resilience Suite, comprising AquaResQ and MobiResQ, across cities in India. 

***** The interview and story development was done by Devki Prabhu and Ganesh, Science Writers at the Indian Institute of Technology Gandhinagar*****