Update app.py

app.py

@@ -8,7 +8,7 @@ from transformers import pipeline
 from config import NASA_API_KEY  # Ensure this file exists with your NASA API Key
 
 # Set up Streamlit UI
-st.set_page_config(page_title="HAL - A NASA ChatBot", page_icon="🚀")
+st.set_page_config(page_title="HAL - NASA ChatBot", page_icon="🚀")
 
 # --- Ensure Session State Variables are Initialized ---
 if "chat_history" not in st.session_state:
@@ -26,7 +26,7 @@ model_id = "mistralai/Mistral-7B-Instruct-v0.3"
 # Initialize sentiment analysis pipeline
 sentiment_analyzer = pipeline("sentiment-analysis")
 
-def get_llm_hf_inference(model_id=model_id, max_new_tokens=128, temperature=0.1):
+def get_llm_hf_inference(model_id=model_id, max_new_tokens=128, temperature=0.7):
     return HuggingFaceEndpoint(
         repo_id=model_id,
         max_new_tokens=max_new_tokens,
@@ -53,14 +53,24 @@ def predict_action(user_text):
     return "general_query"
 
 def generate_follow_up(user_text):
+    """
+    Generates a concise and conversational follow-up question related to the user's input.
+    """
     prompt_text = (
-        f"Based on the user's message: '{user_text}', suggest a natural follow-up question "
-        "to keep the conversation engaging."
+        f"Given the user's question: '{user_text}', generate a single friendly follow-up question. "
+        "Make it short, conversational, and natural—like a human would ask. "
+        "Example: If the user asks 'What is a quark?', respond with something like "
+        "'Would you like to learn about the six types of quarks?' "
+        "Do NOT include phrases like 'A natural follow-up question could be'."
     )
-    hf = get_llm_hf_inference(max_new_tokens=64, temperature=0.7)
+
+    hf = get_llm_hf_inference(max_new_tokens=32, temperature=0.7)
     return hf.invoke(input=prompt_text).strip()
 
 def get_response(system_message, chat_history, user_text, max_new_tokens=256):
+    """
+    Generates HAL's response, making it more conversational and engaging.
+    """
     sentiment = analyze_sentiment(user_text)
     action = predict_action(user_text)
 
@@ -74,20 +84,30 @@ def get_response(system_message, chat_history, user_text, max_new_tokens=256):
         chat_history.append({'role': 'assistant', 'content': follow_up})
         return response, follow_up, chat_history, nasa_url
 
-    hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.1)
+    hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.9)
 
     prompt = PromptTemplate.from_template(
-        "[INST] {system_message}\n\nCurrent Conversation:\n{chat_history}\n\nUser: {user_text}.\n [/INST]\nAI:"
+        (
+            "[INST] {system_message}"
+            "\nCurrent Conversation:\n{chat_history}\n\n"
+            "\nUser: {user_text}.\n [/INST]"
+            "\nAI: Keep responses conversational and engaging. Start with a friendly phrase like "
+            "'Certainly!', 'Of course!', or 'Great question!' before answering."
+            " Keep responses concise but engaging."
+            "\nHAL:"
+        )
     )
+
     chat = prompt | hf.bind(skip_prompt=True) | StrOutputParser(output_key='content')
     response = chat.invoke(input=dict(system_message=system_message, user_text=user_text, chat_history=chat_history))
-    response = response.split("AI:")[-1]
+    response = response.split("HAL:")[-1].strip()
 
     chat_history.append({'role': 'user', 'content': user_text})
     chat_history.append({'role': 'assistant', 'content': response})
 
+    # ✅ Removes unnecessary sentiment apology
     if sentiment == "NEGATIVE":
-        response += "\n😞 I'm sorry to hear that. How can I assist you further?"
+        response += ""  # Keeps response normal instead of adding "I'm sorry to hear that"
 
     follow_up = generate_follow_up(user_text)
     chat_history.append({'role': 'assistant', 'content': follow_up})
@@ -156,14 +176,11 @@ if user_input:
         chat_history=st.session_state.chat_history
     )
 
-    # Display HAL's response
     st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {response}</div>", unsafe_allow_html=True)
 
-    # Display NASA image if available
     if image_url:
         st.image(image_url, caption="NASA Image of the Day")
 
-    # Store follow-up question in session state
     st.session_state.follow_up = follow_up
     st.session_state.response_ready = True  # Enables follow-up response cycle