research / 2026
structure-grounded medical qa
RDF and SPARQL retrieval with claim-level verification for faithful medical question answering.
Python · RDFLib · SPARQL · RAG
Overview
Built a medical question-answering pipeline that retrieves structured evidence from RDF graphs and verifies generated claims against that evidence. I am second author on the associated ACL 2026 SURGeLLM workshop submission.
System path
question → entity resolution → SPARQL retrieval → evidence assembly → answer generation → claim verification
What I built
- Structured retrieval over medical RDF data using SPARQL.
- Deterministic conversion of query results into evidence objects with provenance.
- Claim-level checks that separate supported, unsupported, and conflicting statements.
- Evaluation plumbing for comparing answer faithfulness against retrieved evidence.
Why structure matters
Text retrieval can return semantically related passages without making relationships explicit. RDF triples preserve entities and relations, allowing each generated claim to be checked against a concrete evidence path.
Research status
Submitted to the ACL 2026 SURGeLLM Workshop as “Structure-Grounded Medical QA: RDF Retrieval and Claim-Level Verification for Faithful Answering.”