How I got here

During the COVID lockdowns, I was in class 12 and preparing for JEE. Computer science was one of my subjects, so I had to submit a project, and nobody was sure whether exams would even happen.

So I went to YouTube, found Flask, and started building small websites in Python. I picked whatever seemed interesting at the time. Officially it was for the project. Really, it was a way to clear my head between JEE prep sessions. That habit of building things just to learn how they work never went away.

In 2023, ChatGPT was new and Google had just released Bard (before it became Gemini). I was discussing a project idea with a friend and had done a lot of the research in Bard. Then I hit a wall: there was no way to share that chat with him.

So I built one. Share Chit Chat was a browser extension, backed by a small Flask API and MongoDB, that let you share a Bard conversation with a link. It grew to 25+ weekly active users. That isn’t a big number, but it was the first time strangers used something I made because it solved a problem they also had.

By late 2023, I’d done a handful of hackathons. At first they were an excuse to explore: travel somewhere new, meet new people, see what happens. Then I got hooked on the format. You get 24 or 48 hours, a problem, and a team, and you have to ship something. A deadline that close leaves no room to overthink.

In Smart India Hackathon 2023, our team of six took on a problem statement from NTRO (the National Technical Research Organisation) and built Digital Fortress, a tool that linked dark-web forum activity to Bitcoin transactions. I worked on the data pipeline with my teammates: scraping forums, pulling out Bitcoin addresses, and tracing where the money went. That project is also where I fell for scraping and reverse engineering, working out how a site really works under the hood.

We made the finals but didn’t win. Afterward, our mentors told us we had missed first place by two points. That one hurt the whole team.

We kept going. In March 2024, we placed in the top 10 at DevsHouse with Buildify, a no-code platform for building LLM apps, where I built the engine that turns a drag-and-drop workflow into a running API. A few weeks later, in April, my team won the MongoDB prize at WittyHacks with FinCraze, where I built the Android app and the Telegram chatbot.

Missing first place in 2023 wasn’t the end of that hackathon. Some time later, the organization behind our problem statement reached out to our team about an internship and interviewed us. That’s how I spent the summer of 2024 interning with the Government of India. The use case was sensitive, so the system had to run on an air-gapped network: no external APIs, no cloud, no internet to fall back on.

I worked on it alone, from design to deployment. I built an offline RAG pipeline over the organization’s own data, using ChromaDB and locally hosted LLMs, and designed it so any local model could be swapped in. I also explored knowledge graphs with Neo4j. Building without the cloud taught me a lot, because when you can’t call someone else’s API for any piece, you have to understand how each part works yourself.

One evening I saw a tweet from the founder of Styl, a swipe-to-shop fashion app. They were looking for someone who could do web scraping, and there was an assignment attached.

I already loved scraping from the hackathons, and I had a free evening, so I did the assignment just for fun and emailed it over. That led to an intro call, and soon I was working at Styl part-time while still in college.

When I joined, the team ran scrapers on their own laptops and uploaded CSVs by hand. I built a pipeline that ran all of them on a VM with one click, refreshed every brand on a schedule, and validated the data before it reached the live catalog. In late November, I stepped back to focus on my December exams, and on one more hackathon.

A year after missing by two points, we went back to Smart India Hackathon, again for NTRO’s cybersecurity domain. This time we built Cyber News Monitor, a real-time feed of cyber incidents, and we won. We presented the final solution at IIT Kharagpur.

(One strange tradition: in both Smart India Hackathons, someone on the team had a medical scare in the middle of the event, a different teammate each year. Nothing long-term, but it made both events memorable in ways we didn’t plan.)

It was my twelfth hackathon in two years.

Then Sage AI Labs acquired Styl, and I came along with the acquisition. I’ve been at Sage since January 2025.

The core of my work is getting product data from online retailers. Every store is built differently, so each one needs its own extraction script. I’ve onboarded 55+ retailers by hand: writing the script, validating the data, and plugging it into pipelines that process over 100K products a day.

But we needed to add retailers in bulk, and we’re a small engineering team. Writing every script by hand doesn’t scale. So, as a team, we decided to automate it, and I’m building the AI harness that writes these scripts. It finds how a store serves its products, writes the script, checks samples against the live site, and hands the result to a person for approval. A script that took 1 to 72 hours by hand now takes about 30 minutes to 2 hours, and 165+ of them have been merged into our pipeline so far.

Looking back, the best things I’ve built started from something right in front of me: a missing share button, a manual CSV routine, a retailer queue that keeps growing.

I’m still the person who opens YouTube to figure out how something works. These days I open an AI agent instead, and some days I end up teaching the agent instead of the other way around.