Loading…
Hi, I'm Himanshu. I build AI products, developer tools and open-source systems. This is where I share what I build, what I learn, and resources for the community.
Hey, I'm Himanshu 👋
Tools and systems people install and use.
Local proxy for Claude Code that asks TypeSafe's Jev which model — Haiku, Sonnet or Opus — fits each new turn.
Local-first evaluation framework for RAG systems and AI agents — retrieval, generation and cost metrics behind one CLI and SDK.
Python-first model manager for local LLMs — discover, install, update and verify models across local runtimes from one SDK and CLI.
Event-driven multi-agent system that runs inside GitHub Actions to triage issues, review PRs, repair CI and resolve conflicts.
Short, practical notes and cheat sheets written while building.
What MCP actually standardizes, how the client/server/host roles fit together, and why it matters for connecting LLM agents to tools and data sources.
The core concepts and control-flow patterns in LangGraph — nodes, edges, state, and the conditional routing that makes it different from a plain LangChain chain.
A short reference on how to actually measure whether a retrieval-augmented pipeline is working — retrieval metrics, generation metrics, and the failure modes each one catches.
Where my time goes at the moment.
RAG pipelines end to end — chunking, embeddings, vector search and the evaluation harness that tells you whether any of it improved.
LangGraph workflows, MCP servers and multi-agent systems that call real tools reliably, not just in a demo.
CLIs and proxies that live in the terminal and get used daily — local-first and fast enough to replace the old habit.