Update app.py

app.py

@@ -23,13 +23,19 @@ MODEL = "qwen/qwq-32b:free"
 # ✅ FastAPI app
 app = FastAPI()
 
-# ✅ Context Retriever
-def get_context(query, top_k=5):
+# ✅ Context Retriever with threshold
+def get_context(query, top_k=5, threshold=0.3):
     query_vec = embed_model.encode([query])
     D, I = index.search(np.array(query_vec), top_k)
-    return "\n".join([texts[i] for i in I[0]])
 
-# ✅ Chat Function
+    # Filter based on similarity distance threshold
+    results = [(texts[i], d) for i, d in zip(I[0], D[0]) if d >= threshold]
+    if not results:
+        return ""
+    
+    return "\n".join([text for text, _ in results])
+
+# ✅ Chat Function (RAG-aware + fallback)
 def chat_fn(message, history):
     headers = {
         "Authorization": f"Bearer {API_KEY}",
@@ -38,8 +44,15 @@ def chat_fn(message, history):
 
     context = get_context(message)
 
+    # No valid context? Respond with fallback message
+    if not context.strip():
+        return "❌ Sorry, I can't answer that."
+
     messages = [
-        {"role": "system", "content": "You are a helpful assistant. Use the following context to answer: " + context}
+        {
+            "role": "system",
+            "content": "You are a helpful assistant. Only use the context below to answer. If you can't find an answer in it, reply: 'Sorry, I can't answer that.'\n\nContext:\n" + context
+        }
     ]
 
     for user, assistant in history:
@@ -79,7 +92,7 @@ demo = gr.ChatInterface(
 # ✅ Mount Gradio at root
 app = gr.mount_gradio_app(app, demo, path="/")
 
-# ✅ For local dev (optional on Hugging Face)
+# ✅ For local development
 if __name__ == "__main__":
     nest_asyncio.apply()
     uvicorn.run(app, host="0.0.0.0", port=7860)