Update main.py

main.py

@@ -743,12 +743,20 @@ def generate_with_fallback(messages):
 # ══════════════════════════════════════════════════════════
 # ask_legal
 # ══════════════════════════════════════════════════════════
+import hashlib
+_answer_cache: Dict = {}
+
 def ask_legal_core(question: str) -> dict:
+    cache_key = hashlib.md5(question.strip().encode()).hexdigest()
+    if cache_key in _answer_cache:
+        logger.info("Cache hit!")
+        return _answer_cache[cache_key]
+
     original = question
     question, norm_log = normalize_question(question)
 
     if not is_legal_question_llm(question):
-        return {'answer': OUT_OF_SCOPE_RESPONSE, 'sources': [], 'coverage': 0,
+        result = {'answer': OUT_OF_SCOPE_RESPONSE, 'sources': [], 'coverage': 0,
                 'model': 'out_of_scope', 'normalized': question}
 
     queries    = rewrite_query(question)
@@ -806,7 +814,7 @@ def ask_legal_core(question: str) -> dict:
         coverage   = calculate_coverage(question, final_docs)
 
     if len(final_docs) == 0 or coverage < 0.20:
-        return {'answer': OUT_OF_SCOPE_RESPONSE, 'sources': [], 'coverage': 0,
+        result = {'answer': OUT_OF_SCOPE_RESPONSE, 'sources': [], 'coverage': 0,
                 'model': 'quality_check', 'normalized': question}
 
     context = build_context(final_docs)
@@ -816,15 +824,18 @@ def ask_legal_core(question: str) -> dict:
     ])
 
     if not answer:
-        return {'answer': 'كل الموديلات محجوزة حالياً.', 'sources': [], 'coverage': 0, 'model': ''}
+        result = {'answer': 'كل الموديلات محجوزة حالياً.', 'sources': [], 'coverage': 0, 'model': ''}
 
-    return {
+    result = {
         'answer':     post_process(answer, final_docs),
         'sources':    [{'law': d.metadata.get('law_name',''), 'article': d.metadata.get('article_number','')} for d in final_docs[:3]],
         'coverage':   round(coverage * 100),
         'model':      model_used,
         'normalized': question,
     }
+    if result['model'] not in ['out_of_scope', 'quality_check', '']:
+        _answer_cache[cache_key] = result
+    return result
 
 # ══════════════════════════════════════════════════════════
 # FastAPI
@@ -860,11 +871,11 @@ class QuestionResponse(BaseModel):
 
 @app.get('/')
 def root():
-    return {'name':'Saudi Legal AI','version':'4.2.0','status':'running','docs':'/docs'}
+    result = {'name':'Saudi Legal AI','version':'4.2.0','status':'running','docs':'/docs'}
 
 @app.get('/health')
 def health():
-    return {
+    result = {
         'status':        'healthy',
         'chunks':        vectorstore._collection.count() if vectorstore else 0,
         'groq_models':   [m['name'] for m in ACTIVE_MODELS],
@@ -913,7 +924,7 @@ async def ask(req: QuestionRequest, request: Request):
 
 @app.get('/stats')
 def get_stats():
-    return {
+    result = {
         'total':      stats['total'],
         'success':    stats['success'],
         'blocked':    stats['blocked'],
@@ -926,7 +937,7 @@ def get_laws():
     law_names = sorted(set(
         m.get('law_name','') for m in vectorstore.get()['metadatas'] if m.get('law_name')
     ))
-    return {'laws': law_names, 'total': len(law_names)}
+    result = {'laws': law_names, 'total': len(law_names)}
 
 if __name__ == '__main__':
     import uvicorn