Ankush Sil Sarma

At ZS Associates, I build end-to-end structured and unstructured data pipelines for clinical activity ingestion and protocol load processing across 20,000+ artifacts. My work powers GenAI, analytics, product usage tracking, and knowledge-extraction workflows.

I built a LangGraph multi-agent chatbot that maps 150+ natural-language query patterns to enterprise schemas and enables governed Text-to-SQL through AWS Bedrock, reducing analyst exploration time by approximately 3x. I also engineer retrieval systems using OpenSearch, FAISS, Neptune SPARQL, learning-to-rank, and multi-signal relevance scoring.

I own production OpenSearch indices and Neptune graph systems, design PostgreSQL and MySQL data models, and maintain 100+ REST APIs in a scalable microservices ecosystem. I orchestrate distributed AWS pipelines with MWAA, Airflow, AWS Batch, and EKS.

My research includes AI-powered knowledge extraction, biomedical RAG evaluation, retrieval-augmented self-distillation, and multi-agent debate systems, with work accepted at CVMI 2026 and COMSYS 2026.