Local-first AI developer tooling
Exploring how local models can be discovered, attached to developer workflows and exposed through a Copilot-like extension experience — with GPU/SSH execution as a deployment layer.
The lab / live work
What I’m actually working on.
This page is deliberately more dynamic than a résumé. It shows the current engineering loop: experiments, active repositories and public activity.
Deep learning → LLM engineering
Building the foundations that make later transformer, fine-tuning, inference and evaluation work understandable rather than magical.
Live public stream
GitHub activity.
Loading live activity…
Working rhythm
Build notes, not fake “daily logs.”
Frame the problem
Define what the system must make possible before selecting a model, framework or API.
Get one vertical slice working
Prefer a small end-to-end path over a pile of disconnected notebooks.
Measure the failure
Logs, traces, benchmarks and reproducible experiments turn “it doesn’t work” into an engineering problem.
Document the tradeoffs
The goal is not just code that runs — it is code whose decisions can be defended in an interview or production review.
Put it somewhere real
Deploy, expose, test and iterate. The portfolio should show evidence of shipping, not only screenshots.