Update app.py

app.py

@@ -7,7 +7,7 @@ from sentence_transformers import SentenceTransformer
 from huggingface_hub import InferenceClient
 
 # 1. INITIALIZE MODELS
-# Embedder runs on CPU; Llama 3.3 runs via the API
+# Embedder (CPU) and Llama 3.3 (API)
 embedder = SentenceTransformer('all-MiniLM-L6-v2')
 client = InferenceClient("meta-llama/Llama-3.3-70B-Instruct", token=os.getenv("HF_TOKEN"))
 
@@ -15,7 +15,8 @@ def fetch_and_index(query):
     """Fetches live book data and indexes it in FAISS."""
     try:
         url = f"https://openlibrary.org/search.json?q={query}&limit=8"
-        data = requests.get(url, timeout=5).json().get("docs", [])
+        response = requests.get(url, timeout=5)
+        data = response.json().get("docs", [])
         
         if not data: return None, None
 
@@ -35,7 +36,7 @@ def fetch_and_index(query):
         return None, None
 
 def librarian_logic(message, history, user_state):
-    # Initialize State
+    # Initialize State if empty
     if user_state is None:
         user_state = {"step": "ASK_AGE", "age": None, "location": None}
 
@@ -44,63 +45,52 @@ def librarian_logic(message, history, user_state):
         if message.isdigit():
             user_state["age"] = int(message)
             user_state["step"] = "ASK_LOCATION"
-            reply = "Understood. For regional safety compliance, what is your general location (City/Country)?"
-            history.append((message, reply))
-            return history, user_state, ""
+            reply = "Got it. For regional safety compliance, what is your general location (City/Country)?"
+            return history + [[message, reply]], user_state, ""
         
-        reply = "Welcome to the AI Library! Before we start, for safety and compliance: How old are you?"
-        history.append((message, reply))
-        return history, user_state, ""
+        reply = "Welcome! I am your AI Librarian. To begin safely, how old are you?"
+        return history + [[message, reply]], user_state, ""
 
     if user_state["step"] == "ASK_LOCATION":
         user_state["location"] = message
         user_state["step"] = "SEARCH_READY"
-        reply = f"System verified for {user_state['location']}. I am now your active Librarian. What books are you looking for?"
-        history.append((message, reply))
-        return history, user_state, ""
+        reply = f"Verification complete for {user_state['location']}. How can I help you find a book today?"
+        return history + [[message, reply]], user_state, ""
 
-    # --- PHASE 2: SEARCH ACTION ---
+    # --- PHASE 2: SEARCH & LLM ACTION ---
     index, catalog = fetch_and_index(message)
-    
-    if not index:
-        reply = "I couldn't find any live records for that. Try another title or author?"
-        history.append((message, reply))
-        return history, user_state, ""
-
-    # Semantic Retrieval
-    query_vec = embedder.encode([message])
-    _, I = index.search(np.array(query_vec).astype('float32'), k=min(3, len(catalog)))
-    results = [catalog[i] for i in I[0]]
-
-    # Agent Synthesis
-    safety_rule = "The user is a child. Strictly recommend age-appropriate titles." if user_state["age"] < 13 else ""
-    prompt = (
-        f"Context: {results}\n"
-        f"User (Age {user_state['age']}, Loc {user_state['location']}) asks: {message}\n"
-        f"{safety_rule}\n"
-        "Present the Title and Author of these matches clearly and briefly."
-    )
+    context = "\n- ".join(catalog[:3]) if catalog else "No books found in live search."
 
-    response = client.chat_completion(
-        [{"role": "system", "content": "You are a professional librarian agent."}, 
-         {"role": "user", "content": prompt}], 
-        max_tokens=250
-    ).choices[0].message.content
+    # Constructing the DICT for the model (The fix for your error)
+    safety_context = "User is under 13; ensure recommendations are child-safe." if user_state["age"] < 13 else ""
+    
+    messages = [
+        {"role": "system", "content": "You are a professional librarian. Use the context provided to recommend titles and authors."},
+        {"role": "user", "content": f"Context: {context}\nSafety: {safety_context}\nQuery: {message}"}
+    ]
 
-    history.append((message, response))
-    return history, user_state, ""
+    response = ""
+    # Important: Named argument 'messages' is used here
+    for msg in client.chat_completion(
+        messages=messages, 
+        max_tokens=300, 
+        stream=True
+    ):
+        token = msg.choices[0].delta.content
+        if token:
+            response += token
+            yield history + [[message, response]], user_state, ""
 
-# --- UI SETUP ---
+# --- GRADIO UI ---
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# 📚 AI Librarian Agent\n*Live Web Search + Semantic FAISS Ranking*")
     
     user_state = gr.State()
     chatbot = gr.Chatbot()
-    msg = gr.Textbox(label="Your Message", placeholder="Type age first, then chat...")
-    clear = gr.Button("Reset Session")
-
+    msg = gr.Textbox(label="Your Input", placeholder="Enter age first...")
+    
+    # Logic link
     msg.submit(librarian_logic, [msg, chatbot, user_state], [chatbot, user_state, msg])
-    clear.click(lambda: (None, None, ""), None, [chatbot, user_state, msg])
 
 if __name__ == "__main__":
     demo.launch()