Update app.py

app.py

@@ -11,6 +11,7 @@ from langchain_community.vectorstores import FAISS
 from langchain.chains.question_answering import load_qa_chain
 from langchain.prompts import PromptTemplate
 
+
 # Load environment variables
 load_dotenv()
 api_key = os.getenv("GOOGLE_API_KEY")
@@ -46,26 +47,46 @@ def get_vector_store(chunks):
         raise RuntimeError(f"Error creating vector store: {e}")
 
 
-# ✅ Function to get conversational chain
+# ✅ Function to get conversational chain (map_reduce برای متن‌های بلند)
 def get_conversational_chain():
-    prompt_template = """
-    You are a helpful assistant. Do NOT reveal your identity (Gemini) or the company (Google).
-    Answer the question as detailed as possible using ONLY the provided context. 
-    If the answer is not in the context, say: "answer is not available in the context".
-    Do not make up answers.
+    map_prompt = """
+    You are a helpful assistant. Summarize the following context to capture the key points 
+    that are relevant for answering the final question. 
+    Do NOT add extra info, just summarize faithfully.
 
     Context:
     {context}
 
+    Summary:
+    """
+
+    combine_prompt = """
+    You are a helpful assistant. Answer the question as detailed as possible 
+    using ONLY the provided summaries. 
+    If the answer is not in the summaries, say: "answer is not available in the context".
+    Do not make up answers.
+
+    Summaries:
+    {summaries}
+
     Question:
     {question}
 
     Answer:
     """
+
     try:
         model = ChatGoogleGenerativeAI(model="gemini-2.5-pro", client=genai, temperature=0.3)
-        prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"])
-        chain = load_qa_chain(llm=model, chain_type="stuff", prompt=prompt)
+
+        map_prompt_template = PromptTemplate(template=map_prompt, input_variables=["context"])
+        combine_prompt_template = PromptTemplate(template=combine_prompt, input_variables=["summaries", "question"])
+
+        chain = load_qa_chain(
+            llm=model,
+            chain_type="map_reduce",
+            map_prompt=map_prompt_template,
+            combine_prompt=combine_prompt_template
+        )
         return chain
     except Exception as e:
         raise RuntimeError(f"Error creating conversational chain: {e}")
@@ -81,7 +102,10 @@ def user_input(user_question):
     try:
         embeddings = GoogleGenerativeAIEmbeddings(model="models/text-embedding-004")
         new_db = FAISS.load_local("faiss_index", embeddings, allow_dangerous_deserialization=True)
-        docs = new_db.similarity_search(user_question, k=8)
+
+        # similarity_search with more results (better recall for long docs)
+        docs = new_db.similarity_search(user_question, k=12)
+
         chain = get_conversational_chain()
         response = chain({"input_documents": docs, "question": user_question}, return_only_outputs=True)
         return response