Kunal User

I’m an AI Engineer specializing in production LLM systems: RAG pipelines, LLM fine-tuning (LoRA/PEFT), AI agents, and multimodal search. At Tata Steel, I shipped a real-time computer vision pipeline—deployed a YOLOv8 PPE detection system served via a FastAPI REST API and monitoring live CCTV feeds with zero manual intervention.

I also built an IoU-based PPE-person association engine integrated with a React dashboard, and resolved annotation inconsistencies across 5 datasets to improve performance (domain adaptation raised safety-vest representation from 2.3% to 33.5%, lifting mAP from 0.897 to 0.910). On my projects, I architected a multi-tenant RAG SaaS (DocuMind), implemented NeurIPS 2025 Federated LLM fine-tuning (FedLEASE) using LoRA with a non-IID federation, and built a multimodal agentic response system (RESPOND) with shared vector storage and zero human-in-the-loop operation.