All projects
Part of the CHU project · individual work
Medical document scanner
Turns unstructured Word medical reports into structured patient records, anonymising patients before anything reaches the LLM.
What I did
-
Python pipeline (LLM + RAG): from unstructured Word reports to structured patient records, ready for the main API.
-
Patients are pseudonymised with spaCy NER before anything reaches the LLM, then restored after extraction.
-
Validation and faithfulness checks on every extracted value; command line and Electron desktop app, Gemini or Claude as the LLM, pytest suite.