I build resilient backend systems and contribute to open-source agentic AI frameworks. My focus is on distributed systems, LLM orchestration, and shipping code that solves real problems — not just code that runs on my machine.
I approach engineering as a problem solver first — breaking down ambiguous requirements into clean system designs before writing a single line of code. Whether it's choosing the right concurrency model, designing a fault-tolerant queue, or debugging a race condition in async agent pipelines, I think in systems.
Currently contributing to ML4Sci's DeepLense project and building production-grade backend APIs with Python and LangGraph.
Mesa-LLM — LLM-powered agent simulation framework (800+ ⭐)
- Fixed async event-loop blocking in
astep()usingrun_in_executor→ 60% latency reduction - Resolved JSON serialization crash affecting all agent state exports; added pytest regression suite
- Shipped a 12-agent concurrent simulation demo — merged by 3 maintainers
ML4Sci · DeepLense — GSoC '26 Evaluation
- Trained EfficientNet-B0 on 30K scientific images → 0.99 AUC ROC
- Built a CLI agent with NLP-to-API translation, schema enforcement, and human-in-the-loop confirmation
| Project | What it does | Stack |
|---|---|---|
| ReceiptAgent | AI receipt processing engine — OCR extraction, LLM parsing, fraud detection, async Celery workers | Django LangGraph Celery Redis Docker |
| Intelligent Study Assistant | RAG-based document Q&A and summarization system | Python LangChain ChromaDB |
| Paradigm Classifier | NLP-based programming paradigm detection model | Python scikit-learn |
- 550+ LeetCode — 400+ Medium · 70+ Hard
- Strong in: Dynamic Programming · Graphs · Trees · Binary Search · Sliding Window · Union-Find · Topological Sort
- CodeChef 3-Star — 1600+ rating · 15+ rated contests · Division 2