About

2+ years in full-stack development, and at home across the Gen AI stack — LLM infrastructure, retrieval, and automation that runs in production.

I work out of Howrah, just across the river from Kolkata. Most of what I have built in the last year sits in one of three places: products people can actually open, infrastructure that other developers would use, and automation that runs against live systems.

The through line is boundaries. In Fluid, the interesting part is not that a model generates an interface — it is that the model is held to a closed grammar it cannot argue with, validated before anything is stored. In SlothOps, it is that every automated action has an approval gate that defaults to on. In Project Phoenix, it is choosing regex over inference so a forensic report is reproducible by whoever reads it next.

I write mostly TypeScript and Python. I like problems where the correct answer is a constraint rather than a feature, and I would rather ship a small thing that holds under load than a large one that demos well.

How I work

Most of what I build starts because I wanted the thing to exist and the existing version got some detail wrong — voice notes treated as attachments, personalised interfaces that only scale by hiring people, forensic tools that answer differently on Tuesday.

I care most about the boundaries in a system: what a model is allowed to emit, what a stateless relay is allowed to forget, which actions need a human before they run.

Working out of Howrah, mostly in TypeScript and Python, and reachable below.

How I got here

Self-taught, then formally, and still mostly by building. The repository list is honest about that — there is a run of small experiments and course work underneath the projects on the home page, and I have left them public on purpose.