feat: brouillon de reponse copiable, briefing, balayage mobile, resume de pieces jointes, apprentissage par exemples, sauvegarde quotidienne, vue Aujourd'hui, references PO/AO, correctif reponses vides des modeles de raisonnement
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XGaEeTZLvHXfYPUNWkgYoi
This commit is contained in:
1 parent
f87c56fee7
commit
054a4d64e8
15 files changed
+721
-19
No files matched your search
+108
@@ -0,0 +1,108 @@
|
||||
"""Resume d'une piece jointe a la demande (tt-2026-11-072). Le texte est extrait ici, localement, puis envoye au seul
|
||||
modele autorise via PrivAiTe (pseudonymisation). Ni le contenu, ni le resume ne sont conserves ; refus net si l'IA est
|
||||
desactivee ou PrivAiTe injoignable ; types actifs et fichiers trop gros refuses ; scans sans texte signales."""
|
||||
import base64
|
||||
import io
|
||||
import re
|
||||
import zipfile
|
||||
from typing import Optional
|
||||
from xml.etree import ElementTree
|
||||
|
||||
from fastapi import APIRouter, HTTPException, Query
|
||||
|
||||
from app import ia
|
||||
|
||||
MAX_OCTETS = 8 * 1024 * 1024
|
||||
MAX_TEXTE = 12000
|
||||
TEXTE = {"txt", "csv", "md", "log"}
|
||||
|
||||
|
||||
class SansTexte(Exception):
|
||||
pass
|
||||
|
||||
|
||||
def _ext(nom: str) -> str:
|
||||
return nom.rsplit(".", 1)[-1].lower() if "." in nom else ""
|
||||
|
||||
|
||||
def extraire_texte(nom: str, brut: bytes) -> str:
|
||||
e = _ext(nom)
|
||||
if e in TEXTE:
|
||||
t = brut.decode("utf-8", "replace")
|
||||
elif e == "docx":
|
||||
try:
|
||||
with zipfile.ZipFile(io.BytesIO(brut)) as z:
|
||||
racine = ElementTree.fromstring(z.read("word/document.xml"))
|
||||
except Exception:
|
||||
raise SansTexte("document illisible")
|
||||
t = "\n".join("".join(x.text or "" for x in p.iter("{http://schemas.openxmlformats.org/wordprocessingml/2006/main}t"))
|
||||
for p in racine.iter("{http://schemas.openxmlformats.org/wordprocessingml/2006/main}p"))
|
||||
elif e == "xlsx":
|
||||
try:
|
||||
import openpyxl
|
||||
wb = openpyxl.load_workbook(io.BytesIO(brut), read_only=True, data_only=True)
|
||||
lignes = []
|
||||
for ws in wb.worksheets[:3]:
|
||||
lignes.append("# " + ws.title)
|
||||
for i, row in enumerate(ws.iter_rows(values_only=True)):
|
||||
if i >= 80:
|
||||
break
|
||||
lignes.append(" | ".join("" if c is None else str(c) for c in row))
|
||||
t = "\n".join(lignes)
|
||||
except Exception:
|
||||
raise SansTexte("classeur illisible")
|
||||
elif e == "pdf":
|
||||
try:
|
||||
from pypdf import PdfReader
|
||||
t = "\n".join((p.extract_text() or "") for p in PdfReader(io.BytesIO(brut)).pages[:15])
|
||||
except Exception:
|
||||
raise SansTexte("PDF illisible")
|
||||
else:
|
||||
raise SansTexte("type non pris en charge (PDF, Word, Excel, texte)")
|
||||
t = re.sub(r"[ \t]+", " ", t)
|
||||
t = re.sub(r"\n{3,}", "\n\n", t).strip()
|
||||
if len(t) < 40:
|
||||
raise SansTexte("aucun texte lisible (probablement un scan)")
|
||||
return t[:MAX_TEXTE]
|
||||
|
||||
|
||||
CONSIGNE = ("Tu resumes une piece jointe pour un responsable achats. En francais, 4 a 6 lignes maximum : nature du document, "
|
||||
"points essentiels, montants, quantites, dates et echeances, action attendue s'il y en a une. Ne fais aucune "
|
||||
"supposition : si une information manque, ne l'invente pas. Pas de formule d'introduction.")
|
||||
|
||||
|
||||
def resumer_texte(texte: str) -> str:
|
||||
return ia.appeler_privaite([{"role": "system", "content": CONSIGNE}, {"role": "user", "content": texte}], max_tokens=1200).strip()
|
||||
|
||||
|
||||
def creer_routeur(mirror_client) -> APIRouter:
|
||||
router = APIRouter(prefix="/ia")
|
||||
|
||||
@router.post("/piece")
|
||||
def resumer_piece(id: str = Query(..., min_length=1, max_length=400), nom: str = Query(..., min_length=1, max_length=260)):
|
||||
if not ia.ia_active():
|
||||
raise HTTPException(status_code=409, detail="L'IA est désactivée.")
|
||||
try:
|
||||
res = mirror_client.get_attachment(id, nom)
|
||||
except Exception:
|
||||
raise HTTPException(status_code=502, detail="Pièce jointe indisponible : miroir injoignable.")
|
||||
d = res.get("donnees") if isinstance(res, dict) else None
|
||||
b64 = d.get("contenu_base64") if isinstance(d, dict) else None
|
||||
if not b64:
|
||||
raise HTTPException(status_code=404, detail="Pièce jointe introuvable.")
|
||||
if len(b64) * 3 // 4 > MAX_OCTETS:
|
||||
raise HTTPException(status_code=413, detail="Pièce jointe trop volumineuse pour un résumé (8 Mo maximum).")
|
||||
try:
|
||||
brut = base64.b64decode(b64)
|
||||
except Exception:
|
||||
raise HTTPException(status_code=502, detail="Pièce jointe illisible.")
|
||||
try:
|
||||
texte = extraire_texte(nom, brut)
|
||||
except SansTexte as e:
|
||||
raise HTTPException(status_code=422, detail="Résumé impossible : " + str(e) + ".")
|
||||
try:
|
||||
return {"nom": nom, "resume": resumer_texte(texte), "extrait_chars": len(texte)}
|
||||
except ia.IAIndisponible as e:
|
||||
raise HTTPException(status_code=503, detail="IA indisponible, aucun résumé : " + str(e) + ".")
|
||||
|
||||
return router
|
||||
Reference in new issue
Block a user