- π§ I build AI agents and the systems that keep them running
- β‘ LangGraph, FastAPI, and PyTorch are basically muscle memory
- π§ Linux native. I think in systems, not scripts.
- π Hackathon winner, Codeforces Pupil, and a serial shipper of real projects
- π My rule is simple: build it, break it, ship it, repeat
| Role | Company | What I shipped |
|---|---|---|
| SDE Intern (Jan 2026 to Mar 2026) | Kinzy | Guided-flow architecture serving 5000+ active users, a personalization engine (20+ categories, 100+ flow paths), and a real-time leaderboard handling 1000+ submissions a day |
| AI/ML + Backend Engineer (Oct 2025 to Dec 2025) | Rahtsaa AI | A document extraction pipeline on GCP (1000+ docs a month), backend services that cut processing time by 60%, and LLM-based extraction with automated workflows |
| Product Engineer / Data Scientist (Jul 2025 to Aug 2025) | Liquidmind.ai | A multi-modal RAG system on Azure with 99.5% uptime and parallel pipelines that ran 2.5x faster |
π§ mimir
Gives autonomous agents a memory. They store what happened, recall it later, and stop repeating the same mistakes. Python
βοΈ Kensei
An AutoML and MLOps platform that trains, tunes, and ships ML models as FastAPI services from a single command. Python
π€ QueryCrafter
A multi-agent research system on LangGraph and LangChain. It breaks a question into parts, searches across sources, and writes up the answer. Python
π VectorVault
A semantic search engine that handles embeddings, caching, FAISS search, a FastAPI backend, and a Streamlit UI. Python
A drag-and-drop builder for wiring up AI and data workflows in the browser. JavaScript
- π₯ National Hackathon Winner: feedforward NN for real-time fault detection on a microprocessor, 90%+ accuracy at sub-100ms latency
- π₯ 1st Prize, Innovation Design Thinking Hackathon: CNN neonatal jaundice detector, 92% accuracy across 500+ samples
- π₯ 3rd Place, Hardware Hackathon: Arduino sensor and API automation system
- π₯ 1st Rank, "DISS THAT DSA" Contest
- π» Competitive Programming: Codeforces Pupil (1200+), LeetCode 1624
- π Active across ML, backend, and open source (live stats below)
πΈοΈ "With great power comes great responsibility."
I care more about whether something works in production than how clever it looks in a notebook. Build it, break it, ship it.
CUDA optimization Β· LLM fine-tuning Β· multi-agent architectures Β· high-performance inference Β· open source






