I bridge the gap between complex research-backed models (RAG optimization, multi-agent orchestrations, real-time edge CV) and robust, high-throughput systems (production-grade memory debugging, high-performance database algorithms).
|
Frameworks & APIs: PyTorch, TensorFlow, LangChain, LangGraph, Hugging Face, YOLOv8, Gemini API, Anthropic API Cloud & Infrastructure: AWS (SageMaker, EC2, S3, Lambda), Docker |
Languages: C++, Java, Node.js, SQL, TypeScript Frameworks & Databases: Django, React, Redis, Microservices, PostgreSQL |
| Project | Tech Stack | Impact & Key Metrics |
|---|---|---|
| LLMEval | Python, Gemini API, Anthropic API | LLM-as-judge framework scoring RAG outputs on faithfulness, relevance, hallucination, coherence; validated against human ratings via Cohen's Kappa. |
| UrbanMind | LangGraph, React, TypeScript | Multi-agent system (intent classification, recommendation, rendering nodes, fallback/replanning logic) driving an interactive floor-plan UI. |
| GreenLight Go | YOLOv8, OpenCV, C++ | Real-time ambulance detection with 97.3% mAP @ 30 FPS, INT8-quantized for CPU-only edge deployment. |
Important
During my SDE Internship at OrganicOrbits, I worked on low-level systems profiling and optimization in a high-throughput microservices environment:
- π οΈ Production Memory Leak Resolution: Investigated and debugged a critical memory leak in a high-scale system handling 10,000+ events/minute with sub-100ms p95 latency requirements, stabilizing the production cluster.
- β‘ 40% Processing Time Reduction: Replaced a bottlenecked lookup algorithm with an optimized O(1) data structure, cutting data processing time by 40%.
- π§ͺ Pre-release Quality Safeguards: Penned comprehensive regression and integration tests that caught 3 critical pre-release bugs, preventing potential downtime.
- Haystack (Deepset) (21k+ β): Merged a PR fixing an embedding precision bug.
- traceAI: Merged a PR fixing a critical span-overwrite bug corrupting distributed trace data.
- Codeforces: Expert (Peak Rating: 1769, top 2% globally).
- CodeChef: 3-Star (Peak Rating: 1749, top 0.06% worldwide).
- TVS Credit EPIC Hackathon: Ranked 2nd / 62,000+ teams.
- ChurnZero Hackathon (IIT Kharagpur): Finalist (among 1,300+ teams).