dom 4d49d4e114 feat: grounding CPAM — tags DP/DAS/ANT/COMPL + fuzzy matching CIM-10 + prompt renforcé ...

Cause racine du Tier C : le LLM inventait des tags ([C83.3], [Antécédents])
car _build_tagged_context() ne taguait que bio/img/trt/actes. Le DP, les DAS,
antécédents et complications n'avaient aucun tag citable.

- cpam_context: 4 nouveaux types de tags [DP], [DAS-N], [ANT-N], [COMPL-N]
- cpam_validation: fuzzy matching — résout les refs CIM-10 nues vers le tag contenant ce code
- templates: liste explicite des tags valides, interdiction d'inventer des tags
- tests: 18 nouveaux tests (tags, fuzzy match, grounding DAS/DP)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-02-20 13:56:07 +01:00

config

feat: BIO_NORMALS 33 analytes + interprétations cliniques + cohérence DAS/bio étendue

2026-02-20 11:00:53 +01:00

data

fix: garde-fous qualité Phase 1 — codes invalides et raisonnements vides

2026-02-20 07:53:43 +01:00

docs

feat: architecture multi-modèles LLM + quality engine + benchmark

2026-02-20 00:21:09 +01:00

scripts

feat: parallélisation pipeline --workers N (ThreadPoolExecutor)

2026-02-20 01:30:51 +01:00

src

feat: grounding CPAM — tags DP/DAS/ANT/COMPL + fuzzy matching CIM-10 + prompt renforcé

2026-02-20 13:56:07 +01:00

tests

feat: grounding CPAM — tags DP/DAS/ANT/COMPL + fuzzy matching CIM-10 + prompt renforcé

2026-02-20 13:56:07 +01:00

.env.example

feat: configuration externalisée via .env + audit requirements

2026-02-13 19:46:33 +01:00

.gitignore

feat: configuration externalisée via .env + audit requirements

2026-02-13 19:46:33 +01:00

analyze_pdfs.py

feat: pipeline T2A - anonymisation, extraction CIM-10 et intégration edsnlp

2026-02-10 15:24:12 +01:00

batch_50.sh

feat: architecture multi-modèles LLM + quality engine + benchmark

2026-02-20 00:21:09 +01:00

pyproject.toml

feat: infrastructure — pyproject.toml, requirements-dev, conftest, pytest-cov

2026-02-20 10:06:11 +01:00

rapport_analyse_pdfs.md

feat: pipeline T2A - anonymisation, extraction CIM-10 et intégration edsnlp

2026-02-10 15:24:12 +01:00

requirements-dev.txt

feat: infrastructure — pyproject.toml, requirements-dev, conftest, pytest-cov

2026-02-20 10:06:11 +01:00

requirements.txt

feat: architecture multi-modèles LLM + quality engine + benchmark

2026-02-20 00:21:09 +01:00

run.sh

feat: architecture multi-modèles LLM + quality engine + benchmark

2026-02-20 00:21:09 +01:00

Description

No description provided

203 MiB

Languages

Python 87.3%

HTML 12.3%

Shell 0.4%