Update main.py

main.py

@@ -1,8 +1,9 @@
 import os
-from fastapi import FastAPI, UploadFile, File, HTTPException
+from fastapi import FastAPI, UploadFile, File, HTTPException, Form
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 from groq import Groq
+import base64
 
 app = FastAPI(title="BSTP-Cameroun-AI-Engine")
 
@@ -18,7 +19,7 @@ app.add_middleware(
 GROQ_API_KEY = os.getenv("GROQ_API_KEY")
 
 if not GROQ_API_KEY:
-    print("⚠️ Attention: GROQ_API_KEY n'est pas configurée dans les secrets de l'espace Hugging Face.")
+    print("Attention: GROQ_API_KEY n'est pas configurée dans les secrets de l'espace Hugging Face.")
 
 groq_client = Groq(api_key=GROQ_API_KEY) if GROQ_API_KEY else None
 
@@ -31,12 +32,12 @@ if os.path.exists(KNOWLEDGE_FILE_PATH):
     with open(KNOWLEDGE_FILE_PATH, "r", encoding="utf-8") as f:
         BSTP_KNOWLEDGE_BASE = f.read()
 else:
-    print(f"⚠️ Erreur: Le fichier {KNOWLEDGE_FILE_PATH} est introuvable. Initialisation d'une base vide.")
+    print(f"Erreur: Le fichier {KNOWLEDGE_FILE_PATH} est introuvable. Initialisation d'une base vide.")
     BSTP_KNOWLEDGE_BASE = "Base de connaissances non disponible."
 
-# =========================================================
+# ========================
 # SYSTEM PROMPT EN ANGLAIS
-# =========================================================
+# ========================
 SYSTEM_PROMPT = (
     "ROLE AND MANDATE:\n"
     "You are BSTP-Intellect, the advanced, specialized, and authoritative AI Governance & Sourcing Assistant "
@@ -78,7 +79,6 @@ SYSTEM_PROMPT = (
     "If the query is in English, close naturally with: 'For any further guidance regarding the BSTP ecosystem or the execution of our operational cockpits, I remain entirely at your disposal.'"
 )
 
-# Assemblage final du Prompt et du contexte de la base de connaissances
 FULL_SYSTEM_PROMPT = SYSTEM_PROMPT + "\n\nOFFICIAL BSTP REFERENCE CONTEXT FROM DATABASE:\n" + BSTP_KNOWLEDGE_BASE
 
 class TextRequest(BaseModel):
@@ -106,7 +106,7 @@ async def chat_text(request: TextRequest):
                 {"role": "system", "content": FULL_SYSTEM_PROMPT},
                 {"role": "user", "content": request.text}
             ],
-            temperature=0.3, # Température basse pour maintenir une exactitude maximale
+            temperature=0.3,
             max_tokens=1024
         )
         return {"ai_response": completion.choices[0].message.content}
@@ -122,22 +122,18 @@ async def chat_voice(file: UploadFile = File(...)):
         raise HTTPException(status_code=500, detail="Le moteur d'IA Groq n'est pas configuré.")
     
     try:
-        # Étape A : Transcription de la note vocale industrielle
         transcription = groq_client.audio.transcriptions.create(
             file=(file.filename, await file.read()),
             model="whisper-large-v3",
             response_format="json"
         )
-        
         user_text = transcription.text
-        
         if not user_text or not user_text.strip():
             return {
                 "user_said": "", 
                 "ai_response": "Je n'ai pas pu intercepter de flux audio distinct concernant la BSTP. Pouvez-vous reformuler ? / I could not process the audio instruction. Please try again."
             }
         
-        # Étape B : Soumission du texte transcrit au modèle orienté BSTP
         completion = groq_client.chat.completions.create(
             model="llama-3.3-70b-versatile",
             messages=[
@@ -147,11 +143,68 @@ async def chat_voice(file: UploadFile = File(...)):
             temperature=0.3,
             max_tokens=1024
         )
+        return {"user_said": user_text, "ai_response": completion.choices[0].message.content}
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Erreur du pipeline vocal Groq BSTP : {str(e)}")
+
+# ===================================
+# ROUTE 3 : AUDIT DOCUMENTAIRE VISION 
+# ===================================
+@app.post("/api/document-audit")
+async def audit_document(file: UploadFile = File(...), document_type: str = Form(...)):
+    """
+    document_type doit être l'un des suivants: 'rccm', 'niu', 'cnps', 'attestation_fiscale'
+    """
+    if not groq_client:
+        raise HTTPException(status_code=500, detail="Le moteur d'IA Groq n'est pas configuré.")
+    
+    valid_types = ['rccm', 'niu', 'cnps', 'attestation_fiscale']
+    if document_type.lower() not in valid_types:
+        raise HTTPException(status_code=400, detail=f"Type de document invalide. Choisissez parmi : {valid_types}")
+    
+    try:
+        # Lecture de l'image et encodage en Base64 pour l'API Vision de Groq
+        image_bytes = await file.read()
+        base64_image = base64.b64encode(image_bytes).decode('utf-8')
+        
+        # Construction du prompt d'analyse d'image ciblé sur la base de connaissances
+        vision_prompt = (
+            f"You are the document verification submodule of BSTP-Intellect.\n"
+            f"The user has uploaded an image that is claimed to be a '{document_type.upper()}'.\n"
+            f"Analyze this image carefully. Perform OCR to extract key administrative data, then cross-reference "
+            f"the visual text with the official 'SME PROFILING DOCUMENT COMPLIANCE CRITERIA' in our knowledge base.\n\n"
+            f"Provide a structured JSON response in the identical language of the system with the following keys:\n"
+            f"1. 'is_valid': boolean (true if it matches the expected document type and criteria, false otherwise).\n"
+            f"2. 'extracted_info': a short text summarizing the key numbers, dates, or corporate names identified.\n"
+            f"3. 'compliance_report': a detailed administrative explanation of why the document is accepted or rejected based on Cameroon regulations.\n"
+            f"Do not return any conversational text around the JSON, return ONLY a valid JSON object."
+        )
+        response = groq_client.chat.completions.create(
+            model="llama-3.2-11b-vision-preview", 
+            messages=[
+                {
+                    "role": "system",
+                    "content": f"{BSTP_KNOWLEDGE_BASE}\n\nYou must strictly output valid JSON structures."
+                },
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": vision_prompt},
+                        {
+                            "type": "image_url",
+                            "image_url": {
+                                "url": f"data:image/jpeg;base64,{base64_image}"
+                            }
+                        }
+                    ]
+                }
+            ],
+            temperature=0.1, 
+            response_format={"type": "json_object"} 
+        )
         
-        return {
-            "user_said": user_text,
-            "ai_response": completion.choices[0].message.content
-        }
+        import json
+        return json.loads(response.choices[0].message.content)
         
     except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Erreur du pipeline vocal Groq BSTP : {str(e)}")
+        raise HTTPException(status_code=500, detail=f"Erreur lors de l'analyse OCR/Vision par Groq : {str(e)}")