This is the third part of me and my entrepreneurial dream.
Over time, I noticed one thing: we're doing product building, working on niche ideas, taking ownership—and everything seems fine. But the key thing about entrepreneurship is owning all these things completely. The idea has to be ours. If I have to build a venture, build my own company, it has to be mine. Not part of anybody else's.
The Inner Conflict
That's where the inner conflict started. At Mu Sigma, I was saying things like "running a company"—getting deals done, doing all these things. At Captain Fresh, I called myself an entrepreneur in residence. Everything seemed fine. But did I have entire ownership of things?
That's what gives you founder mode. You get things done, you own it, you fight to sell, you get momentum, you get traction. But that was something that kept missing. That's what stops the fulfillment of the journey.
I feel like 2026 should give an answer to that. I shouldn't just be under somebody's dream or somebody's mission, being a cog in the main engine or wheel.
This is something I want to reflect on. That's what I'm figuring out. That's where Pracha Labs comes in—to bring that fulfillment. Find my own problem, build it, get it to market, see how the traction looks and how people are really interested in it. I'm aware of this, but still working through it.
The Informatica Phase
Going back to the building phase. At Informatica, in the initial days, it was all about doing applied research—continual training, developing our own conversation mining from customer support conversations so we can develop what people now call agent skills. Claude actually has this now.
There were projects on customer behavior modeling, causal inference—different aspects. But the real product building was about the AI plus observability stack. There were backend engineers working on connectors, fetching observability data—metrics, traces, logs from software systems like data engineering systems.
The company had a lot of customers running a lot of jobs. We were getting traces and data, but we wanted to bring insights for quick troubleshooting when jobs failed, or for optimization. I was building an application—not a platform, more of a microservice. I was collaborating with many engineers to understand how things work.
Since it's AI, we weren't just doing AI engineering. We were also running small experiments on custom models, learning from the conversations themselves. And the design aspect—that's when I started reading more about design itself. What makes people adopt a product?
A product has utility, it has a use case, but adoption was something we struggled with. It wasn't compatible with engineers' day-to-day usability. They wanted something easy to adapt, and our solution was demanding multiple windows. That was tiring for engineers.
So we evolved the solution from engineer-facing to immersed in workflows—making it very automated delivery to customers and engineers. That's how it went.
The Stealth Startup: Building from Zero
Parallelly, I was asked to help a company in stealth mode. I don't want to reveal the name because they're still growing, still figuring out their customers.
They were doing years of document processing—consider it a BPO or KPO company with a lot of manual processing. Obviously, they wanted to build a document AI company. They had the customer base, the use case, knew the business and domain. They wanted AI adoption.
They had tried partnering with IBM and local vendors but were skeptical about the stories these vendors sold. I was asked to help them understand whether these vendors were approaching the problem correctly.
From the beginning, I was driving conversations about what they want as a solution. I was designing how the AI solution should look. With IBM, they had engineers and partners building something. But ultimately, the level of accuracy and precision required for document processing wasn't met by IBM's models or their partner engineers.
That's when they asked me to build their own AI team.
Building an AI Team from Scratch
I helped them build the AI team, but we didn't hire experienced AI engineers. Instead, I started training people who were working on React and complete freshers who had never worked on anything. Based on training, small projects, and continuous monitoring, we built the first AI team.
Over six to ten months, that team is now in very good shape. They own end-to-end data processing—data preparation to model deployment, building agents, doing validations. Over those months, I was helping them work with UI/UX developers, designers, validating different vendor platforms, and training models.
For a company that's small and in a budding state, we trained our own models to build a unique value proposition—a small medical-focused model we developed ourselves. Over time, the company built a dataset of around 50,000 annotated documents. We have a team of seven to eight people who have learned and built hands-on experience on AI engineering.
The team has the capacity to train models. I was able to take a very rough state—no team at all—and over time build structure and organization: data processing, operations, machine learning engineers, AI and agentic workflow development, integration with CRM and other platforms. Parallelly, we were dealing with what could be done with IBM partners.
It was cool. I could see the team is now on autopilot mode. They can figure things out. I was happy to give them shape, get validation, help them develop these assets and products.
The Real Question
But again, if I go back and see—we've done a lot of things, but what do I have in my hand now? I helped all these people build and get shape and validation. I helped them develop assets and products.
But where's mine?
That's why the inner battle is all about: when am I going to start alone? That's the need of the hour.
On one side, I want to build a company. I want to get into venture building, building very good AI products for businesses. I'm very much interested in that—not consumer products. There are a lot of generic AI consumer products with good market revenue. Tapping into that isn't something I find interesting. But I'm very interested in how larger enterprises can benefit on very specific problem spaces.
On the other side is my curiosity and interest in learning more depth of machine learning and data science. I believe that's the foundation I want to build on. I don't want to be like many people who develop products based on what they're good at—like being good at backend or quickly developing software with coding. They develop mail assistants, editing software, something, something.
The Founder Mindset Shift
One thing I have to keep in mind: when we say we're building a company, it doesn't mean we build everything ourselves. We run things, we make things happen—not necessarily write all the code. We don't always have to write the frontend, write the backend. That's not always the case.
Sometimes that might cloud the mind: "Oh, we don't know how to deploy the model, but we know how to come up with a unique model." That shouldn't always be the blocker. At least we can learn how to build and deploy and optimize the model. But we cannot compel ourselves to learn frontend as well. That's there.
On one side, I'm focusing on foundations of machine learning, finding where I can find a better level playing ground in ML itself. Can it be music generation? Video generation? Behavioral models? Last semester, I was helping friends on their project ideas, guiding them on how simulations would work. In the past, I worked on agent-based models, which is very relatable these days.
But I need to find my very specific niche to build that company. Machine learning cannot be the only niche—we need very specific focus. Is it going to be foundation models? Prediction models? Causal inference? Simulations?
Convergence: From Ocean to Focus
At least now I'm converging from the very large scope of data science and AI to frontier machine learning that should bring value to humans—human-centric problems rather than operation-centric problems.
There are problems like supply chain optimization, production optimization, inventories—a lot of operation-side work. I'm finding that less interesting now. Over time, as I read more about humans and human-centric problems, I feel it should be more about AI and humans. Even AI agents and their coexistence with humans—this is the place I'm trying to converge, rather than wandering around the huge ocean.
We have that experience, and sometimes learnings might help. But now is the time for convergence.
Sometimes I think: if I'm not starting a company, let's say I get into some other company or laboratory and lead their AI or ML practice—applied ML or applied AI practice for enterprises. Leading engagements, owning products under that. But still, other things are going out of my interest. I want to fully focus on these frontier aspects with humans in mind. There are multiple things like alignment-related issues. I'm also trying to apply for MATS.
The Vector Sum
I hope in the upcoming days—I feel like 2026 is a great start—I'll be able to converge into machine learning, humans, and a commoditizable product for industries. That's what I believe, and that's what I'm working on.
Sometimes I need to keep things very simple, at a surface level. I've seen many people build simple applications—where people can talk to AI, or plot different graphs from internet traffic, or an API that extracts data from TikTok. That's how many Y Combinator companies I've observed work.
But what I'm grounding my learnings on is different. I'm learning about complexity science from Geoffrey West. I'm reading about human values, human behavior from Daniel Kahneman. All these things give me a different perspective about problems and how to approach them—rather than directly jumping and saying, "Let's put some prompts, let's put some agents, let's just solve the problem."
My problem-solving DNA needs a mutation to become a founder DNA. That's something in a different league. I'm aware of it and addressing it—how I can channelize what I've learned, my belief systems, my interested areas towards the destiny of building a venture.
Being in New York and being in the US is giving me hope that I'll be able to do this. This is the right time, as I talked about in my five-year plan. It's all going to be there. I'm not just doing my masters and going to start my career fresh. No. I'm doing my masters while also trying to get started with my venture building journey.
It's a long way to go. I hope all my learnings, my failures, my mistakes, and my aspirations guide the next steps.
Finding Clarity
I feel very clear about where I'm overcomplicating things or where I'm deviating from the typical path of building companies. It's actually like a vector sum. One direction is the dream of building a venture. Another aspiration is interest towards science, towards AI aspects.
These perspectives and aspirations have to meet—or maybe they won't meet—but they have to have a very good vector sum so it comes along in a fruitful way.
So watch this space. I'm going to continuously talk about what I'm trying to build. Check out Pracha Labs for some of my free experiment side of things. And on my LinkedIn, I continuously talk about it.
Let's see where this goes.
The journey continues.