At CitiusTech, I own the end-to-end LLM pipeline for a GenAI clinical scribe that turns physician-patient conversations into structured SOAP notes. I re-architected an unreliable single-prompt summarizer into a multi-stage pipeline and helped cut physician charting time by 60%, with 98% note-agreement in the pilot.
I also developed a citation-aware, agentic RAG assistant over clinical guidelines, using groundedness checks and guardrails to improve answer relevance by 35%. Before CitiusTech, I worked on reinforcement-learning simulations and precision-agriculture systems at Hexaware Technologies, and federated learning and demand forecasting at Encode Testers.
My project work includes LegalSummon, a citation-aware legal RAG service, and UrbanAir, a real-time air-quality forecasting project. Across these projects and roles, I've worked across generative AI, machine learning, and production MLOps on cloud platforms.
