At George Institute for Global Health, I developed a production multi-turn RAG chatbot for pregnancy-related clinical queries. Its retrieval pipelines grounded answers in medical guidelines, with approximately 90% accuracy.
I also built analytics to track session duration and query-classification accuracy, and engineered reusable prompts and ingestion of unstructured documents into vector databases.
At IIT Bombay, I researched privacy-preserving machine learning for medical imaging and developed deep-learning MRI harmonization models for multi-site data. This work was published at ISBI and IEEE ICKG.
At IIT Kharagpur, I proposed an intent-based metric for legal judgment summaries and built a graph-based summarizer that outperformed LetSum and BERT baselines. My research also includes transformer-based lexical complexity prediction published at ACL SemEval 2021.
