Applied GenAI: Evidence Extraction & Cross-Paper Synthesis
LLM-based pipeline for extracting structured data from clinical trial literature and synthesising evidence across papers.
An LLM-based pipeline that extracts structured data from clinical trial literature using constrained, schema-driven prompts on a locally hosted model, achieving a 92% successful extraction rate across a 600+ paper pilot corpus. Extended with a cross-paper synthesis layer combining hybrid dense + BM25 retrieval, cross-encoder reranking, and map-reduce narrative synthesis to produce citation-grounded answers across multiple papers.
Tech stack: Python, LLM prompt engineering, local model hosting, hybrid retrieval (dense + BM25), cross-encoder reranking, map-reduce synthesis
GitHub: [GITHUB LINK TBD]