Final-year CS undergrad at MNNIT Allahabad building AI systems at the intersection of reinforcement learning, LLMs, and production engineering.
Meta × Hugging Face Hackathon 2026 — Top 800 of 31,000+ teams globally (Top 2.6%)
Built Pulse-ER: a Python-based RL training environment for AI-driven trauma care
Sankalp Innovation Challenge 2026 — National Finalist, Team Lead
Built Governance Memory AI: multi-agent civic platform for crisis detection and dispatch
National AI & Innovation Summit, MNNIT Allahabad
RL training environment for AI-driven trauma patient management
- Fine-tuned Qwen2.5-3B-Instruct with GRPO + LoRA (rank 16) for ATLS-aligned protocol sequencing
- Engineered reward shaping pipelines achieving 8.33 expert reward vs −17.15 random baseline (25× improvement)
- Stack: Python · TRL · GRPO · LoRA · FastAPI · Docker · Pulse Physiology Engine
Scalable multi-agent civic platform for complaint clustering and crisis detection
- Architected end-to-end RESTful API pipeline for multilingual complaint intake and dispatch
- Built semantic clustering pipeline with SentenceTransformers + DBSCAN for real-time crisis detection
- Stack: Python · FastAPI · PostgreSQL · SentenceTransformers · scikit-learn · Streamlit
Core: Python · PyTorch · FastAPI · Docker
LLMs & Training: Hugging Face Transformers · TRL · GRPO · LoRA · PEFT · Qwen2.5
ML: scikit-learn · Pandas · NumPy · DBSCAN · SentenceTransformers
Databases: PostgreSQL · MongoDB · SQLite
Infra: Git · Docker · GCP · REST APIs
Interests: Reinforcement Learning · LLM Fine-Tuning · Agentic AI · RAG · NLP
Currently building agentic AI systems and exploring LLM fine-tuning for domain-specific reasoning. Open to AI/ML Engineer and Research roles — 2027 graduate.