Hi, I'm
Anant Mishra.
I build production AI systems and the full-stack apps around them. Most of my work lives at the intersection of LLM agents, retrieval pipelines and scalable backends — the parts of an AI product that decide whether it actually works for real users.
Lately I've been deep in multi-agent RAG platforms with avatar interfaces — the kind of system where retrieval design, streaming and observability matter more than the model itself. Before that, motion-driven marketing sites tuned for speed and SEO. On the side I publish open source on NPM and ship MCP servers for creative tools.
I care about three things: latency, observability, and not lying to users. If your product is going to talk to humans, the words it says had better be grounded in your data and shipped fast enough to feel alive.
- Based
- Noida, IN
- Focus
- AI agents & RAG systems
- CodeChef
- 1400 rating
- LeetCode
- 200+ solved
- Open Source
- lightweight-expression-detector / 109+ wkly installs
Ship the boring path first
Most AI demos break on edge cases nobody planned for. I build the unhappy paths before adding agents.
Filter, don't trust
RAG is 90% retrieval design — chunking, metadata, filters. The model is the easy part.
Streaming UX or it didn't happen
If the user is staring at a spinner for an LLM call, it feels broken. I default to streamed responses.
Logs > guesses
Sentry, structured logs, tracing on every agent call. Production AI without observability is theater.
Open.source
ctxrelay
Stop re-explaining your project every time you switch AI agents.
A zero-dependency CLI that maintains a git-diffable .aicontext/ folder in your repo — a compact, token-budgeted map of your codebase and project state. Switch between Claude Code, Codex, Cursor or a fresh session and the agent picks up where the last one left off, without re-reading the whole tree.
Lightweight Expression Detector
Real-time facial expression detection. Zero dependencies. 109+ weekly NPM installs.
A from-scratch JavaScript package that runs facial expression detection in browser and Node.js with no external deps. Targets builders who want emotion signals without dragging a 50MB ML library along.
The Stack
UPTIME: 99.97