---

👋 About

I build machine-learning systems that run live, not just in notebooks: streaming pipelines, agent frameworks you can replay and audit, and dashboards that make the output usable. B.Tech student at VIT.

I report results the way they came out. The drone project below missed its own 0.88 mAP target, and the README says so.

🚀 Flagship projects

	Project	What it is	Evidence
🛡️	RT-GIDS	Real-time network intrusion detection: XGBoost detector, FastAPI backend, Next.js dashboard with a 3D threat globe and SHAP explainability	33 backend tests, TypeScript typecheck, production build and Docker build on every push
🧭	Deterministic Agent Orchestration	Multi-agent pipeline (decomposer, retriever, critic, synthesizer) where every run produces a replayable trace and a SHA-256 execution hash	123 tests; replay compare and validate endpoints; a React replay/diff view
🔍	Real-Time Fake Job Detector	Streaming pipeline that flags fraudulent job postings as they arrive: Kafka, Spark, PostgreSQL, FastAPI, Next.js	F1 75.6%, precision 82.3%, recall 70.0% on data with 4.8% fraud, with the threshold tuned on a validation split
🔥	Drone Fire Detection with RAG	YOLO edge detection feeding a FAISS-backed RAG pipeline that drafts cited response plans for wildfire responders	[email protected] 0.733 (the 0.88 target was not met), about 73 ms detect-to-plan with a mock LLM, about 198 messages/s across 10 simulated drones

Also: RetailSense Lite , retail demand forecasting with a Prophet, XGBoost and LightGBM ensemble, anomaly detection and a Streamlit dashboard.

🧰 Tech stack

---

_{Open to ML / AI engineering internships and collaborations. Reach out on LinkedIn or by email.}