Add 899999999

rag_service.py

@@ -203,7 +203,10 @@ async def search_docs(query: str, config: RunnableConfig) -> str:
     
     try:
         # Search in the specified knowledge base
+        import time
+        t0 = time.time()
         docs = await lancedb_service.similarity_search(query, userid, knowledge_base)
+        t1 = time.time()
         if docs:
             # Advanced extractive summarization using NLTK
             try:
@@ -212,25 +215,23 @@ async def search_docs(query: str, config: RunnableConfig) -> str:
                 from nltk.tokenize import sent_tokenize
             except ImportError:
                 sent_tokenize = lambda x: x.split('.')
-
+            t2 = time.time()
             # Embedding-based chunk selection
             try:
                 from sentence_transformers import SentenceTransformer
                 embedder = SentenceTransformer('all-MiniLM-L6-v2')
             except ImportError:
                 embedder = None
-
+            t3 = time.time()
             def select_best_chunk(chunks, query):
                 if not embedder or not chunks:
                     return chunks[0] if chunks else ""
                 chunk_embeddings = embedder.encode(chunks)
                 query_embedding = embedder.encode([query])[0]
-                # Cosine similarity
                 import numpy as np
                 scores = [np.dot(chunk_emb, query_embedding)/(np.linalg.norm(chunk_emb)*np.linalg.norm(query_embedding)) for chunk_emb in chunk_embeddings]
                 best_idx = int(np.argmax(scores))
                 return chunks[best_idx]
-
             def extractive_summary(text, max_sentences=3):
                 sentences = sent_tokenize(text)
                 keywords = query.lower().split()
@@ -238,24 +239,28 @@ async def search_docs(query: str, config: RunnableConfig) -> str:
                 if scored:
                     return ' '.join(scored[:max_sentences])
                 return ' '.join(sentences[:max_sentences])
-
+            t4 = time.time()
             compressed_contexts = []
             for doc in docs:
-                # Chunking: split by paragraphs or headings
                 if hasattr(doc, 'chunks') and doc.chunks:
                     best_chunk = select_best_chunk(doc.chunks, query)
                     summary = extractive_summary(best_chunk)
                 else:
-                    # Fallback: split by paragraphs
                     paragraphs = doc.page_content.split('\n\n')
                     best_chunk = select_best_chunk(paragraphs, query) if paragraphs else doc.page_content
                     summary = extractive_summary(best_chunk)
                 compressed_contexts.append(summary)
-
+            t5 = time.time()
             context = "\n\n".join(compressed_contexts)
+            t6 = time.time()
+            import logging
+            logging.info(f"[Latency] Document search: {t1-t0:.3f}s, NLTK setup: {t2-t1:.3f}s, Embedding setup: {t3-t2:.3f}s, Function setup: {t4-t3:.3f}s, Chunking/summarization: {t5-t4:.3f}s, Context join: {t6-t5:.3f}s, Total: {t6-t0:.3f}s")
             return f"📄 Found {len(docs)} relevant documents (chunked & summarized):\n\n{context}"
         else:
             context = ""
+            t7 = time.time()
+            import logging
+            logging.info(f"[Latency] Document search: {t1-t0:.3f}s, No docs found, Total: {t7-t0:.3f}s")
             return "No relevant documents found in the knowledge base."
     except Exception as e:
         logger.error(f"❌ Error searching documents: {e}")