- 🧠 I build AI agents and the systems that keep them running
- ⚡ LangGraph, FastAPI, and PyTorch are basically muscle memory
- 🐧 Linux native. I think in systems, not scripts.
- 🏆 Hackathon winner, Codeforces Pupil, and a serial shipper of real projects
- 🚀 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 |
🧠 mimir
Gives autonomous agents a memory. They store what happened, recall it later, and stop repeating the same mistakes. Python
⚙️ Kensei
An AutoML and MLOps platform that trains, tunes, and ships ML models as FastAPI services from a single command. Python
A multi-agent research system on LangGraph and LangChain. It breaks a question into parts, searches across sources, and writes up the answer. Python
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
Languages
AI / ML
Backend & Data
Cloud & DevOps
Tools
- 🥇 National Hackathon Winner: feedforward NN for real-time fault detection on a microprocessor, 90%+ accuracy at sub-100ms latency
- 🥇 1st Prize, Innovation Design Thinking Hackathon: CNN neonatal jaundice detector, 92% accuracy across 500+ samples
- 🥉 3rd Place, Hardware Hackathon: Arduino sensor and API automation system
- 🥇 1st Rank, "DISS THAT DSA" Contest
- 💻 Competitive Programming: Codeforces Pupil (1200+), LeetCode 1624
- 📈 Active across ML, backend, and open source (live stats below)
🕸️ "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 · LLM fine-tuning · multi-agent architectures · high-performance inference · open source
