feat: configuration externalisée via .env + audit requirements

- Externalise 13 variables de config via python-dotenv (chemins PDF,
  modèles Ollama/embedding/NER, FINESS, seuils) avec défauts identiques
- Centralise EMBEDDING_MODEL dans config.py (était hardcodé en 3 endroits)
- Ajoute .env.example documenté et .env au .gitignore
- Ajoute openpyxl et pandas manquants au requirements.txt
- Ajoute data/referentiels au mkdir de run.sh

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

.env.example

@@ -0,0 +1,22 @@
+# === Référentiels PDF (chemins absolus vers les PDFs ATIH) ===
+# T2A_CIM10_PDF=/chemin/vers/cim-10-fr.pdf
+# T2A_GUIDE_METHODO_PDF=/chemin/vers/guide_methodo_mco.pdf
+# T2A_CCAM_PDF=/chemin/vers/ccam_descriptive.pdf
+
+# === Ollama ===
+# OLLAMA_URL=http://localhost:11434
+# OLLAMA_MODEL=gemma3:12b
+# OLLAMA_TIMEOUT=120
+# OLLAMA_MAX_PARALLEL=2
+
+# === Modèles IA ===
+# T2A_EMBEDDING_MODEL=dangvantuan/sentence-camembert-large
+# T2A_NER_MODEL=Jean-Baptiste/camembert-ner
+# T2A_NER_THRESHOLD=0.80
+
+# === Établissement ===
+# T2A_FINESS=000000000
+# T2A_NUM_UM=0000
+
+# === Anonymisation ===
+# T2A_KEEP_ESTABLISHMENT=True

.gitignore

@@ -16,5 +16,8 @@ data/
 *.xls
 *.xlsx
 
+# Configuration locale
+.env
+
 # IDE / outils
 .claude/

requirements.txt

@@ -11,3 +11,6 @@ faiss-cpu>=1.7.0
 sentence-transformers>=2.2.0
 requests>=2.28.0
 flask>=3.0.0
+python-dotenv>=1.0.0
+openpyxl>=3.0.0
+pandas>=2.0.0

run.sh

@@ -27,7 +27,7 @@ else
 fi
 
 # Créer les répertoires nécessaires
-mkdir -p input output/anonymized output/structured output/reports data/rag_index
+mkdir -p input output/anonymized output/structured output/reports data/rag_index data/referentiels
 
 echo ""
 echo "✨ Application prête !"

src/config.py

@@ -2,11 +2,15 @@
 
 from __future__ import annotations
 
+import os
 from pathlib import Path
 from typing import Optional
 
+from dotenv import load_dotenv
 from pydantic import BaseModel, Field
 
+load_dotenv()
+
 
 # --- Chemins ---
 
@@ -23,24 +27,24 @@ for d in (INPUT_DIR, ANONYMIZED_DIR, STRUCTURED_DIR, REPORTS_DIR):
 
 # --- Configuration anonymisation ---
 
-KEEP_ESTABLISHMENT_NAME = True
+KEEP_ESTABLISHMENT_NAME = os.environ.get("T2A_KEEP_ESTABLISHMENT", "True").lower() in ("true", "1", "yes")
-NER_MODEL = "Jean-Baptiste/camembert-ner"
+NER_MODEL = os.environ.get("T2A_NER_MODEL", "Jean-Baptiste/camembert-ner")
-NER_CONFIDENCE_THRESHOLD = 0.80
+NER_CONFIDENCE_THRESHOLD = float(os.environ.get("T2A_NER_THRESHOLD", "0.80"))
 
 
 # --- Configuration Ollama ---
 
-OLLAMA_URL = "http://localhost:11434"
+OLLAMA_URL = os.environ.get("OLLAMA_URL", "http://localhost:11434")
-OLLAMA_MODEL = "gemma3:12b"
+OLLAMA_MODEL = os.environ.get("OLLAMA_MODEL", "gemma3:12b")
-OLLAMA_TIMEOUT = 120
+OLLAMA_TIMEOUT = int(os.environ.get("OLLAMA_TIMEOUT", "120"))
 OLLAMA_CACHE_PATH = BASE_DIR / "data" / "ollama_cache.json"
-OLLAMA_MAX_PARALLEL = 2
+OLLAMA_MAX_PARALLEL = int(os.environ.get("OLLAMA_MAX_PARALLEL", "2"))
 
 
 # --- Configuration RUM / établissement ---
 
-FINESS = "000000000"
+FINESS = os.environ.get("T2A_FINESS", "000000000")
-NUM_UM = "0000"
+NUM_UM = os.environ.get("T2A_NUM_UM", "0000")
 
 
 # --- Configuration RAG ---
@@ -52,9 +56,13 @@ ALLOWED_EXTENSIONS = {".pdf", ".csv", ".xlsx", ".xls", ".txt"}
 CIM10_DICT_PATH = BASE_DIR / "data" / "cim10_dict.json"
 CIM10_SUPPLEMENTS_PATH = BASE_DIR / "data" / "cim10_supplements.json"
 CCAM_DICT_PATH = BASE_DIR / "data" / "ccam_dict.json"
-CIM10_PDF = Path("/home/dom/ai/aivanov_CIM/cim-10-fr_2026_a_usage_pmsi_version_provisoire_111225.pdf")
+CIM10_PDF = Path(os.environ.get("T2A_CIM10_PDF", "/home/dom/ai/aivanov_CIM/cim-10-fr_2026_a_usage_pmsi_version_provisoire_111225.pdf"))
-GUIDE_METHODO_PDF = Path("/home/dom/ai/aivanov_CIM/guide_methodo_mco_2026_version_provisoire.pdf")
+GUIDE_METHODO_PDF = Path(os.environ.get("T2A_GUIDE_METHODO_PDF", "/home/dom/ai/aivanov_CIM/guide_methodo_mco_2026_version_provisoire.pdf"))
-CCAM_PDF = Path("/home/dom/ai/aivanov_CIM/actualisation_ccam_descriptive_a_usage_pmsi_v4_2025.pdf")
+CCAM_PDF = Path(os.environ.get("T2A_CCAM_PDF", "/home/dom/ai/aivanov_CIM/actualisation_ccam_descriptive_a_usage_pmsi_v4_2025.pdf"))
+
+# --- Modèle d'embedding ---
+
+EMBEDDING_MODEL = os.environ.get("T2A_EMBEDDING_MODEL", "dangvantuan/sentence-camembert-large")
 
 
 # --- Modèles de données CIM-10 ---

src/medical/rag_index.py

@@ -11,7 +11,7 @@ from typing import Optional
 
 import pdfplumber
 
-from ..config import RAG_INDEX_DIR, CIM10_PDF, GUIDE_METHODO_PDF, CCAM_PDF, CCAM_DICT_PATH, REFERENTIELS_DIR
+from ..config import RAG_INDEX_DIR, CIM10_PDF, GUIDE_METHODO_PDF, CCAM_PDF, CCAM_DICT_PATH, REFERENTIELS_DIR, EMBEDDING_MODEL
 
 logger = logging.getLogger(__name__)
 
@@ -426,8 +426,8 @@ def build_index(force: bool = False) -> None:
     # Embeddings — GPU si disponible
     import torch
     _device = "cuda" if torch.cuda.is_available() else "cpu"
-    logger.info("Chargement du modèle d'embedding dangvantuan/sentence-camembert-large (%s)...", _device)
+    logger.info("Chargement du modèle d'embedding %s (%s)...", EMBEDDING_MODEL, _device)
-    model = SentenceTransformer("dangvantuan/sentence-camembert-large", device=_device)
+    model = SentenceTransformer(EMBEDDING_MODEL, device=_device)
     model.max_seq_length = 512  # CamemBERT max position embeddings
 
     texts = [c.text[:2000] for c in all_chunks]  # Tronquer les chunks trop longs

src/medical/rag_search.py

@@ -8,6 +8,7 @@ from concurrent.futures import ThreadPoolExecutor, as_completed
 from ..config import (
     ActeCCAM, Diagnostic, DossierMedical, RAGSource,
     OLLAMA_CACHE_PATH, OLLAMA_MAX_PARALLEL, OLLAMA_MODEL,
+    EMBEDDING_MODEL,
 )
 from .cim10_dict import normalize_code, validate_code as cim10_validate
 from .cim10_extractor import BIO_NORMALS
@@ -36,12 +37,12 @@ def _get_embed_model():
         _device = "cuda" if torch.cuda.is_available() else "cpu"
         try:
             logger.info("Chargement du modèle d'embedding (%s)...", _device)
-            _embed_model = SentenceTransformer("dangvantuan/sentence-camembert-large", device=_device)
+            _embed_model = SentenceTransformer(EMBEDDING_MODEL, device=_device)
         except torch.OutOfMemoryError:
             if _device == "cuda":
                 logger.warning("CUDA OOM pour l'embedding — fallback CPU")
                 torch.cuda.empty_cache()
-                _embed_model = SentenceTransformer("dangvantuan/sentence-camembert-large", device="cpu")
+                _embed_model = SentenceTransformer(EMBEDDING_MODEL, device="cpu")
             else:
                 raise
         _embed_model.max_seq_length = 512