B.Tech ECE @ BIT Mesra
- Pre-final year B.Tech, Electronics & Communication Engineering, BIT Mesra
- Former SWE Intern @ NETSCOUT, where I built a production LLM routing gateway
- Interested in LLM infrastructure: gateways, routing, agent orchestration and reliable AI systems
- Contributing to open source, starting with LiteLLM
| Role | What I built |
|---|---|
| SWE Intern, NETSCOUT | Production LLM routing gateway in FastAPI and asyncio, with circuit breakers, rate limiting and multi-strategy load balancing across LLM providers (proprietary) |
| Project | Description | Highlights |
|---|---|---|
| LLM Routing Gateway internship project, NETSCOUT (proprietary) |
Production gateway that routes requests across multiple LLM providers | FastAPI, asyncio, circuit breakers, rate limiting, multi-strategy load balancing |
| Murmur fork of mightbeanshuu/murmur |
Event-driven AI agent orchestrator: a planner builds a task graph, worker agents run in parallel, validators request revisions, a synthesizer produces the answer | Next.js, Go, PostgreSQL, Redis, Temporal |
| Marg (Smart India Hackathon 2026) 4-person team project, I owned the backend and the ML workflow |
City-wide multi-camera vehicle tracking and re-identification for Indian roads, fusing plate OCR, visual Re-ID and a spatio-temporal feasibility gate | YOLOv8 + ByteTrack, OSNet Re-ID, EasyOCR, FastAPI + SQLite + FAISS, React + TypeScript + Leaflet. Cleared the BIT Mesra internal round |
| UPI Fraud Detection | End-to-end fraud detection system on the PaySim dataset (6.3M transactions) | XGBoost, SMOTE for class imbalance, SHAP explanations, FastAPI + React, Dockerized |
| Medical QA with LoRA | Fine-tuned LLaMA-3.2-1B-Instruct on MedQuAD with LoRA / PEFT on a Colab T4 | ROUGE-L 0.180 → 0.195, BERTScore-F1 0.839 → 0.841 |
| Project | Contribution | Status |
|---|---|---|
| LiteLLM | fix(responses): attribute a mid-stream 429 to the failing deployment: fixed a race where a mid-stream rate limit cooled down the healthy fallback deployment instead of the one that failed | In review |
Languages
Backend & Infra
ML & LLMs
Frontend
Tools
