Jaikishan User

I build production AI systems at L&T, focusing on agentic orchestration, RAG pipelines, and industrial computer vision. In my current role, I lead architecture, evaluation, and end-to-end delivery in Python, turning messy workflows into reliable systems.

With a 4-agent supervisor-worker setup in LangGraph, I cut material certificate verification time from 80–90s to 20–25s and automated cross-plant certificate workflows across 6 sites. I also improved data reliability by tying supplier certificates to SAP master data through a Knowledge Graph.

I make these systems safer and easier to trust by putting observability in place—structured logging, tracing, and agentic evaluation using precision and faithfulness benchmarks against SAP ground truth. I’ve also addressed agentic security risks using Azure AD SSO and RBAC.

Beyond agents, I delivered a production RAG system for 1,000+ learners using semantic chunking, ChromaDB, and a Knowledge Graph for PDF/CSV/SQL queries. Earlier, I co-designed a synthetic data pipeline in Unreal Engine and helped drive YOLO/SSD/CNN training, while building FastAPI and Apache Spark pipelines for telemetry and document preprocessing.