For VisionForgeX, I built the LLM reasoning layer for industrial quality assurance, with fallbacks and validation that rejects root causes citing unknown evidence. I also contributed telemetry anomaly detection and a dashboard with audit reports and automated tests.
I'm co-developing RepairRAG, which traces poisoned answers to source documents and rewrites only affected sentences; its pilot cut attack success from 100% to 6.3%. I’ve also built AI services and applications with FastAPI, LangGraph, and retrieval-augmented generation, alongside ML and full-stack work in my internships.

