Knowledge GraphsShipped
EHR-Clinical-Assistant
Does graph retrieval beat SQL for clinical Q&A?
About
A thesis experiment rigorously comparing graph-based retrieval, SQL, and LLM-only baselines on clinical question answering, across both Claude and open-source Ollama models. 80-question evaluation harness covering 5 question types, plus a doctor-facing chat UI with an interactive patient knowledge graph.
Status
Full evaluation harness, retrieval strategies, and chat UI are complete. Results written up as part of the thesis.
Stack
TypeScriptKuzuPostgresClaudeOllamaSynthea