Upload app.py

app.py

@@ -878,7 +878,11 @@ async def process_api(
 
 
 @app.post("/search/")
-async def search_api(query: str = Form(...), job_id: str = Form(...)):
+async def search_api(
+    query: str = Form(...), 
+    job_id: str = Form(...),
+    turbo_mode: str = Form("false")  # Modo rápido: skip query expansion, menos candidatos
+):
     """
     ENDPOINT DE BUSCA (RAG Híbrido) com Monitoramento de Latência
     """
@@ -887,6 +891,11 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
         raise HTTPException(status_code=404, detail="Job ID não encontrado.")
 
     try:
+        # Parse turbo_mode
+        turbo_mode_bool = turbo_mode.lower() in ("true", "1", "yes")
+        if turbo_mode_bool:
+            logging.info("TURBO MODE ATIVADO - Skip query expansion, menos candidatos")
+        
         model = load_retriever()
         reranker = load_reranker()
         
@@ -896,9 +905,12 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
         bm25_index = cached_data.get("bm25_index")
         
         # ==================================================================
-        # FASE 0: QUERY EXPANSION (Melhora recall)
+        # FASE 0: QUERY EXPANSION (Melhora recall) - Skip em turbo mode
         # ==================================================================
-        expanded_query = expand_query(query)
+        if turbo_mode_bool:
+            expanded_query = query  # Turbo: usa query original (economiza ~5-10s)
+        else:
+            expanded_query = expand_query(query)  # Qualidade: expande query
         
         # ==================================================================
         # FASE 1: HYBRID SEARCH (FAISS Semântico + BM25 Lexical)
@@ -907,7 +919,8 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
         query_embedding = model.encode([expanded_query], convert_to_numpy=True)
         query_normalized = query_embedding / np.linalg.norm(query_embedding, axis=1, keepdims=True)
         
-        top_k_retrieval = min(100, len(df))  # Aumentado para hybrid search
+        # Turbo: menos candidatos = reranking mais rápido
+        top_k_retrieval = min(30 if turbo_mode_bool else 100, len(df))
         
         # --- 1A: Busca Semântica (FAISS) ---
         semantic_scores = {}
@@ -998,7 +1011,7 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
             reverse=True
         )
 
-        final_top_k = 10  # Aumentado de 5 para 10 para melhor contexto
+        final_top_k = 5 if turbo_mode_bool else 10  # Turbo: menos contexto
         final_results = []
         context_parts = []