Projects
All projects
PythonPyseriniRAGLocal LLMsNext.jsSupabaseREST APIs

BioEvidence AI: Evidence-Grounded Biomedical Question Answering

A full-stack biomedical QA application with independently inspectable retrieval, reranking, generation, and citation-validation stages.

2026-08-01 · 1 min read
Live demo
The BioEvidence Examples area also includes detailed TREC BioGen questions, retrieved evidence, generated answers, and citations. Read the connected TREC system case study →
The interaction model
queryretrievalrankcontextevidenceanswer

Follow the path from query to answer.

The interface keeps retrieval, ranking, context, evidence, and the final answer visible as distinct parts of one workflow. The glyph stays intentionally general because that interaction pattern extends beyond a single benchmark or dataset.

Overview

BioEvidence turns a biomedical question into an evidence-grounded answer while keeping retrieval, reranking, generation, and citation validation independently inspectable. Its Examples area makes complete question-to-answer workflows easy to explore without reducing the product to a single benchmark.

System design

  • Pyserini/Lucene retrieval with reranking and evidence selection
  • Local LLM generation followed by citation validation
  • Python backend integrating PubMed/NCBI, Europe PMC, and ClinicalTrials.gov APIs
  • Supabase authentication and application data
  • Next.js interface for live questions, saved examples, and evidence inspection

What was built

  • Inspectable, stage-separated query-to-citation pipeline
  • User-facing Next.js application for live answers and saved example inspection
  • Authentication and application data through Supabase

Result

The result is a practical biomedical QA product with inspectable evidence, API-backed sources, and a complete user-facing workflow from question to citation.