APIs & system architecture
FastAPI services designed around your data model, not a tutorial. Clean service layers, typed contracts, background jobs that retry.
- FastAPI
- SQLAlchemy
- Pydantic v2
- Celery
Python Backend & AI engineer · Kolkata, IN
I build the parts of the product that can't go down.
Python and FastAPI systems that carry multiple tenants, real-time traffic, and LLM features — designed, deployed on AWS, and kept running after launch. Three years of shipping, not prototyping.
200 OK
Every layer above is something I've built, broken, and fixed in production. Tap one.
What I do
Four things, done properly.
LLM features that survive users
Retrieval over your own documents, structured extraction from messy input, and agents that call real tools — with evals from day one.
- RAG
- pgvector
- Function calling
- Evals
Chat, presence & video
WebSocket systems that deliver reliably across every worker, plus self-hosted video meetings on your own domain.
- WebSockets
- Redis pub/sub
- Jitsi
- Docker
Deployment & keeping it up
From an empty EC2 box to a running product: nginx, SSL, systemd, secrets in a vault, and logs that say what actually broke.
- AWS
- nginx
- systemd
- Monitoring
Selected work
Systems in production, with the numbers.
Multi-tenant SaaS platform
A single FastAPI codebase serving isolated customers on their own subdomains, each with a separate database. Tenant resolution happens in middleware and is cached in Redis, so routing costs almost nothing per request.
Celery · Redis · AWS
Real-time chat and self-hosted video
Rebuilt WebSocket delivery to route through Redis pub/sub so messages reach the recipient no matter which worker they're on — fixing a bug where delivery only worked one time in four.
Jitsi · Docker · nginx
Available now
Got a backend that keeps falling over?
Send me what you're building and what's currently in your way. I'll reply within a day with an honest read on whether I'm the right fit.