classMahakisore:
role="Full-Stack Developer"but_mostly="I build with AI"# LLMs · RAG · agents · a bit of ML/DLalso="I ship apps when an idea won't let go"currently="going deeper on agentic systems"defhow_i_work(self):
return ("curious by nature — I'd ""rather understand a thing ""than just use it") |  |
An AI you can actually debate. It runs full Asian & British Parliamentary rounds end-to-end — a FastAPI reasoning engine, a Go real-time gateway, and a RAG pipeline that scores arguments the way a real adjudicator would. Deployed on AWS.
Python · FastAPI · Go · LangChain · pgvector · Redis · Docker · AWS
A finance tracker you can just talk to. Snap a receipt or say it out loud and it turns the mess into clean, structured data — then answers questions about where your money actually goes, in plain English.
FastAPI · LangChain · Gemini · Supabase · Next.js
A benchmark that pits plain LLMs against Vector RAG against GraphRAG on hard, multi-hop biomedical questions — and shows exactly where graph traversal quietly wins and similarity search falls apart.
Python · FastAPI · TigerGraph · Qdrant · LlamaIndex · Docker
Shrinking GPT-2 by ~45% with Kronecker decomposition, then putting back the error I threw away as a tiny sparse correction — a small idea of my own that cut reconstruction error by 47% over the baseline.
Python · PyTorch · NumPy · MATLAB
I like to actually understand the things I build — not just wire them together until they work, but get why they work. Then I build. The whole point, for me, is impact: software that solves a real problem someone actually has. Everything else is just practice. |  understand it first — then build it. |
I'm always up for a good problem, a build worth shipping, or just a conversation.