Dipankar Sarkar

I build AI-native infrastructure, mostly in Rust. Drop-in accelerators for the Python AI stack, local LLM inference, agent tooling, and a safer C. I ship in the open and write about how it works.

Joined on Jun 30, 2026 https://github.com/dipankar

Dipankar Sarkar

Jul 21

Running five coding agents in parallel is easy. Not losing their state when one crashes is the hard part

#ai #rust #agents #opensource

5 min read

Want to connect with Dipankar Sarkar?

Create an account to connect with Dipankar Sarkar. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Dipankar Sarkar

Jul 19

AI agents that browse the web need a fleet of isolated browsers, here is a brokerless scheduler for it

#python #ai #agents #opensource

5 min read

Dipankar Sarkar

Jul 17

A key-value store where the query language is Lua, and you can build RAG inside it

#ai #python #llm #opensource

5 min read

Dipankar Sarkar

Jul 15

A package.lock for the prompts hiding in your codebase

#llm #python #opensource #ai

5 min read

Dipankar Sarkar

Jul 13

I threw 750 autonomous LLM exploit attempts at a $10k sandbox bounty. Zero escapes.

#ai #agents #opensource #llm

4 min read

Dipankar Sarkar

Jul 11

737x faster LangGraph checkpoints, and the case where Rust lost

#rust #python #ai #llm

5 min read

Dipankar Sarkar

Jul 9

Why I built a C-like language for agent-written code

#rust #ai #programming #compilers

5 min read

Dipankar Sarkar

Jul 7

Learning how LLMs actually work by building 18 of them in Zig

#zig #ai #llm #opensource

5 min read

Dipankar Sarkar

Jul 5

Sandboxing untrusted agent code with gVisor costs ~200ms per cold start. Blocking syscalls instead of emulating them costs ~8ms

#ai #agents #devops #opensource

5 min read

Dipankar Sarkar

Jul 2

We let an AI agent hit a database 1034 times. Text-to-SQL ran 23 unsafe ops. The policy layer ran zero

#ai #python #llm #opensource

4 min read

Dipankar Sarkar

Jul 1

I put a Rust layer under LiteLLM. Here is where it actually helped (and where it did not)

#rust #python #ai #performance

3 min read

DEV Community

Dipankar Sarkar

Badges

2 Week Community Wellness Streak

1 Week Community Wellness Streak

Writing Debut

Skills/Languages

Currently hacking on

Available for

Running five coding agents in parallel is easy. Not losing their state when one crashes is the hard part

Want to connect with Dipankar Sarkar?

AI agents that browse the web need a fleet of isolated browsers, here is a brokerless scheduler for it

A key-value store where the query language is Lua, and you can build RAG inside it

A package.lock for the prompts hiding in your codebase

I threw 750 autonomous LLM exploit attempts at a $10k sandbox bounty. Zero escapes.

737x faster LangGraph checkpoints, and the case where Rust lost

Why I built a C-like language for agent-written code

Learning how LLMs actually work by building 18 of them in Zig

Sandboxing untrusted agent code with gVisor costs ~200ms per cold start. Blocking syscalls instead of emulating them costs ~8ms

We let an AI agent hit a database 1034 times. Text-to-SQL ran 23 unsafe ops. The policy layer ran zero

I put a Rust layer under LiteLLM. Here is where it actually helped (and where it did not)