About me

I have always been the kid who was keen about learning new things. I still remember taking apart random things just to see how they worked. Sometimes that meant taking apart my father’s razor, purely to find out what was going on inside it.

I also remember the day we had a computer installed at our home. Back then it was a luxury, and to use one you went to a computer cafe. Speaking of which: I do remember nicking 500 bucks from my parents’ pocket and handing all of it over to the computer cafe, so I could go, use it for a while, and leave when I was done. At 20 bucks an hour I enjoyed quite a few of those sessions before getting my ass busted, and that was when my parents decided it would be better to just get one at home.

At the cafe I only ever thought about playing games. But once the computer was at home and the kid fever to play games washed away, I was fascinated all over again by how things worked. How can my keyboard inputs show up on my screen? That was by far the most fascinating one, because the more I explored, the more I realised this one device could do so much. I could play games, watch videos, solve complex calculations, all of it possible through a computer. Ever since then I had this dream of getting into computer science and working in this domain.

I was also the type of person who liked to build stuff. A castle out of the sofa, a science experiment for the school fair. I like to paint too, because I get to create something and then see the result. This is something I have noticed about myself quite a lot, and the same thing has kept showing up in my career choices.

In my bachelors we had the option to choose a minor, and I decided I wanted to learn more about mobile app development. Learning Java and resolving SDK issues was fun and hair loss rendering. During this time my discrete mathematics teacher called my mother and advised her to guide me towards a career in AI. This was back in 2019, when the craze of LLMs and agentic workflows was not a thing yet, and I will be honest, I did not find it that interesting at the time. Probably because I wanted to get away from the math, lol. Sure, I was good at the math, which is why my professor saw some potential and wanted me to switch paths, but back then I was more drunk on the idea of building my own apps and publishing them on the Play Store and the App Store.

My first job had nothing to do with any of that. I got an opportunity to work as a Business Analyst at a SaaS based company called LeadSquared. It was an unexpected stint in my life, and I very much thank the gods I got to experience it. Despite not being a long stint, I learned a lot. We were obviously taught speaking skills in the professional enhancement classes in our bachelors, but doing it in real life is a different thing. Conversing with stakeholders, handling timelines and deadlines, coordinating with cross functional teams, presenting results, all of these are skills I was able to hone at LeadSquared.

LeadSquared had an amazing get right into it culture, at least from my personal experience. I shadowed a team member who had been there for some time to understand how the company operates, and once I was confident about the process I was assigned my own projects from the get go. Each of these projects was meant to be taken by a business analyst from requirement gathering to product handover, and then extra support for a limited period after that. I really liked working this way. I was able to lead multiple projects, one of which I am particularly proud of, because I finished it well ahead of the deadline without compromising on quality. That implementation resulted in a net 30% revenue increase for the client, and I was even awarded my company’s Spotlight Award.

The role still had that building aspect and the impact aspect, but it lacked the technical aspect. It required technical understanding, but it did not require much coding. And I like to code. So even though I was doing well on the job, I was not really satisfied, and I wanted to shift my path to something more technical.

While I was working as a business analyst, I had a friend in the engineering team, on the backend side, who would create our custom reports. During one of our conversations he got to explaining what he actually did to get a request done. It was for a custom report that pulled data from an external plugin the clients used to extract data. The whole time he was explaining, I was simply thinking, “it would be so nice to have such a role”. That technical aspect I was missing in my job is eventually what moved me to Amsterdam to pursue my masters in Big Data Engineering.

The course structure there was exactly to my liking. You don’t just take exams and get scored to test your knowledge, you also get the chance to showcase it with what you have built. Plus we had full flexibility on the courses we wanted to pick, so most of mine were of that structure. This habit of testing my knowledge by creating projects has just continued over time, and now any time I find a new topic or a new tool I find interesting, I build a small random project around it.

One of the few advantages of being chronically online is that you get all types of news pretty quickly. Any new development, you hear about it very soon. Such was the case with LLMs. Because of the interest my professor had planted, I was already a bit aware of GPT-2, BERT and transformers. I knew of these things but never had the opportunity to properly delve into them, and the independent nature of choosing my own courses during my masters let me do both: strengthen my big data fundamentals to the next level, and combine ML models and LLMs with big data tools to see how they interact.

I was quite fascinated by LLMs and the prospect of what they could and could not do. ChatGPT was pushing boundaries, so I decided I wanted to test one of those boundaries. The thing was, ChatGPT was being used a lot in classrooms. Most students used it to get quick answers, but some also used it to learn new things. I wanted to see if an LLM can teach you better than traditional methods can.

The whole process of getting to the model I wanted was also quite fun. I did not have access to a dedicated GPU on a server, so I was fine tuning on my own device, and even with PEFT LoRA it took a total of 3 days of runtime. Getting there meant properly understanding the things I had only ever read about: how transformers actually attend to context, how the context window caps what the model can hold on to, how tokenization quietly decides what a prompt costs you, and where prompt injection can break something you thought was solid. Then, when building the chatbot, I needed a single source of truth it would stick to instead of improvising around, so I implemented RAG over the curriculum content and retrieved from it through the scaffolded teaching. The results are on the thesis page.

Getting so much exposure to the same thing in different formats made me realise how much more there is to this domain of data. How much depth it has, how many complexities sit in it, and how fun it is to solve those complex questions. Which leads me to where I am today: an engineer who builds pipelines that ingest big data, store it, and serve it out to different verticals that each need their own slice of it. My projects are where I keep proving that to myself.

Why this role? Because it does everything. It lets me build things, it has an impact on the user, and most of all, it is technical.

That is me. If you want to talk shop, or you have a role where I would get to build this stuff every day, I am on the contact tab.