About

I am an AI engineer working on genai, rag and agentic systems. Most of what I have built is agentic and retrieval-based: pipelines that take source material, do something useful with a model, and then have to keep doing it every day without anyone watching.

The part I care about is the gap between a demo that works once and a system that holds up with real users on it. A retrieval pipeline that returns something plausible in a notebook is not the same artefact as one that has been answering for a year, and almost all of the interesting engineering lives in the difference.

I want to go deeper on the model layer. My experience is orchestration and retrieval rather than fine-tuning, and I am looking for work where I can grow into evaluation and model training instead of only building around APIs.

By rebuilding it badly first. I worked through Karpathy’s Zero to Hero series, writing an autograd engine and a character-level language model by hand, and I wrote gradient descent in bare numpy before letting an optimiser do it. Following someone else’s course is not a project and I would not claim it as one, but it is how I stop treating a layer as a black box.

The neural network in C is the one I did unaided, and it is on this site because the interesting part was not writing it. It would not converge, and I spent days certain my backpropagation was wrong. The actual fault was a CSV loader reading a pandas index column as the label, which is to say the one part of the pipeline I never thought to question.

A small team where I own real surface area and ship every week, and where I am close enough to users to see what is actually broken. I would rather be responsible for a feature end to end than one of six people on it.

Open to relocation, requires visa sponsorshipImmediate joiner, open to relocate within India.