Himanshu Kumar
AI engineer building agentic systems, LLM infrastructure and developer tools. Founder of OpenAgentHQ.
My core interest lies in designing systems, not just models — retrieval, pipelines and deployment. I learn by building, and ship it in the open.
- PyPI downloads
- 7,000+
- Merged PRs
- 13+
- Eval metrics
- 18+
- B.E. CSE
- 2022–26
Selected work
Systems I built end to end — retrieval pipelines, evaluation tooling and CLIs that people install and use.
AI Commit
Generates intelligent Git commit messages using local Ollama AI models. Free, private, and works offline!
- CLI
- LLM
- Privacy
- Ollama
RAG-based AI Application
End-to-end RAG pipeline with LangChain, ChromaDB & Sentence Transformers for semantic search and LLM-powered Q&A
- RAG
- NLP
- LLM
- ChromaDB
AutoML Studio
End-to-end ML platform that automates EDA, preprocessing, training, and evaluation without writing code
- AutoML
- ML
- Platform
- No-Code
run-git
Lightweight Python CLI tool that simplifies Git workflows — stage, commit, and push with a single command
- Git
- CLI
- Python
- PyPI
mcp-web-search
Production-ready MCP server for web search with FastAPI, JSON-RPC, and tested protocol compatibility
- FastAPI
- MCP
- SSE
- AI Systems
ML Mastery
Structured Machine Learning learning repository with hands-on notebooks, practical examples, and quick reference guides
- Machine Learning
- Education
- Notebooks
- Python
OpenAgent Eval
Local-first evaluation framework for RAG systems and AI Agents — 18+ metrics, CLI + SDK, framework-agnostic
- AI Evaluation
- RAG
- CLI
- Open Source
Repositories
Pulled live from the GitHub API.
Repositories are not loading right now — browse them on GitHub.
Open source
Work merged into other people's codebases — where the review bar is set by someone else.
Stack
Tools I have shipped something real with. Highlighted entries are the ones I reach for first.
Languages
- Python
- C++
- TypeScript
- SQL
LLM & agents
- LangChain
- LangGraph
- MCP
- Groq
- Gemini
- Ollama
Retrieval & ML
- FAISS
- ChromaDB
- Sentence Transformers
- Scikit-learn
- RAG evaluation
Backend & serving
- FastAPI
- Server-Sent Events
- Celery
- Redis
- Streamlit
Infrastructure
- Docker
- GitHub Actions
- Git
- Vercel
Building in public
Live from the GitHub API — no screenshots, no cached badges.
- Public repos
- Stars earned
- Forks
- Followers
Language distribution
About
CS engineering student at BEU Patna and AI engineer. I build in the open and treat every project as a system, not a notebook.
I started where most people do — tutorials. What changed things was shipping: putting a CLI on PyPI, watching strangers install it, and discovering how much of the work is everything around the model.
That is the throughline of everything here. RAGNOVA taught me that a retrieval pipeline is only as good as its evaluation, which became openagent-eval. Wanting local, private commit messages became run-git and AI Commit. Wanting agents to reach live information became mcp-web-search.
I run OpenAgentHQ as the home for that work — agentic systems, LLM infrastructure and developer tooling, all public, all reviewable. I am currently looking for an AI/ML or GenAI internship where the problems are real and the feedback loop is short.
Retrieval systems
RAG pipelines end to end — chunking, embeddings, vector search and the evaluation harness that tells you whether any of it actually improved.
Agentic infrastructure
LangGraph workflows and MCP servers: the protocol plumbing that lets models call real tools reliably rather than in a demo.
Developer tooling
CLIs that live in a terminal and get used daily — local-first, offline-capable, and fast enough that nobody reaches for the old command.
Writing
Notes on what I learn while building — the details that were not in any tutorial.
Posts
Shorter takes on AI/ML, open source and developer tooling, shared as I go.
Background
Where the formal training sits alongside what I keep pulling at on my own time.
Education
Bihar Engineering University (BEU)
In progressB.E. Computer Science & Engineering
Where my attention goes
- Machine learning & evaluation
- Generative AI & LLM applications
- AI-powered developer tools
- Data analysis & backend systems
Certifications
Verifiable credentials — each one links to its issuer where a public verification page exists.
Claude Code in Action
Anthropic
Get Started with Databricks for Generative AI
Databricks
Introduction to Generative AI
Simplilearn (SkillUp) — powered by Google Cloud
Programming with Python
Internshala Trainings
Get in touch
Open to AI/ML and GenAI internships, open source collaboration, or a conversation about agent infrastructure.