hey, I'm Sawradip 👋

I build things at the intersection of AI infrastructure and language. Usually things that don't exist yet.

Lately I've been obsessed with how much a single person can ship when they stop context-switching and just let the tools rip. Claude Code, Codex, the whole agentic coding stack. RunAgent was largely built this way. What used to take a team now fits in one person's flow.

Currently shipping RunAgent, a serverless runtime for AI agents using Firecracker microVMs. Sub-second cold starts, full isolation, runs LangGraph/CrewAI/LlamaIndex agents like they're serverless functions.

Before that: built the first Bengali LLM (7B params, ~35B tokens) because it didn't exist. Also spent a year at Composio generating the tool layer that most AI agents quietly depend on.

Some glimpse from RunAgent:

Some projects shipped from Intelsense:

at Intelsense AI I was CTO. built these because Bengali had no real AI infrastructure.

research side: Google Scholar · 5 papers · 65+ citations · GSoC 2023 Intel OpenVINO · IEEE VIP Cup 🥇 2021

python typescript rust go · pytorch huggingface langgraph · firecracker docker k8s · claude code codex