At Pi-labs, I architected and deployed a natural-language intelligence platform for CDR/IPDR/TDR data covering 100M+ telecom records. Its self-correcting Text-to-SQL agent reduced query failures by ~80% and cut investigation turnaround from 15 minutes to ~6 minutes.
I also engineered SMITH Server, a hot-swappable prompt and AI-skills registry, and built safeguards that recorded zero query-escape incidents across 10K+ generated queries. My work spans LLM fine-tuning, agentic systems, and production AI infrastructure; I also co-authored AngelBot, a published RAG-grounded conversational AI for maternal healthcare.
