Enhance XAI transparency and user experience: Added real-time conversation display during model interactions, limited model exchanges to 3 rounds, improved response quality and clarity, and refined summary generation. Integrated spinner feedback for better UI responsiveness.

app.py

@@ -38,21 +38,21 @@ def generate_engineer_response(user_text, tokenizer, model):
     """
     prompt = f"""
     User text: {user_text}
-    Provide a technical approach or solution.
+    Provide a technical approach or solution that directly addresses the problem. Ensure your response is actionable and concise.
     """
     inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True)
     outputs = model.generate(
         inputs["input_ids"],
         attention_mask=inputs["attention_mask"],
-        max_new_tokens=100,  # Generate up to 100 new tokens
+        max_new_tokens=80,  # Generate up to 80 new tokens
         temperature=0.7,
         do_sample=True,
-        top_p=0.9,
-        repetition_penalty=1.3,
+        top_p=0.85,
+        repetition_penalty=1.2,
         no_repeat_ngram_size=2,
         pad_token_id=tokenizer.pad_token_id
     )
-    explanation = f"Generated based on user input: {user_text}"
+    explanation = f"Engineer response based on user input: '{user_text}'"
     return tokenizer.decode(outputs[0], skip_special_tokens=True), explanation
 
 def generate_analyst_response(user_text, engineer_output, tokenizer, model):
@@ -62,31 +62,31 @@ def generate_analyst_response(user_text, engineer_output, tokenizer, model):
     prompt = f"""
 Engineer provided the following: {engineer_output}
 
-Provide an approach or solution from a data-centric perspective.
+Based on this, provide an actionable data-driven approach or solution to complement the engineer's perspective.
 """
     inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True)
     outputs = model.generate(
         inputs["input_ids"],
         attention_mask=inputs["attention_mask"],
-        max_new_tokens=100,  # Generate up to 100 new tokens
+        max_new_tokens=80,  # Generate up to 80 new tokens
         temperature=0.7,
         do_sample=True,
-        top_p=0.9,
-        repetition_penalty=1.3,
+        top_p=0.85,
+        repetition_penalty=1.2,
         no_repeat_ngram_size=2,
         pad_token_id=tokenizer.pad_token_id
     )
-    explanation = f"Generated based on Engineer's output: {engineer_output}"
+    explanation = f"Analyst response based on Engineer's output: '{engineer_output}'"
     return tokenizer.decode(outputs[0], skip_special_tokens=True), explanation
 
 def summarize_conversation(conversation):
     """
     Summarize the entire conversation to produce a comprehensive plan.
     """
-    summary = "Here is the summarized plan based on the discussion:\n"
+    summary = "**Final Plan:**\n"
     for speaker, text in conversation:
         if speaker != "User":
-            summary += f"- {speaker}: {text}\n"
+            summary += f"- **{speaker}:** {text}\n"
     return summary
 
 ##############################################################################
@@ -109,48 +109,23 @@ if st.button("Start/Continue Conversation"):
         st.session_state.conversation.append(("User", user_text))
 
         # Engineer generates a response
-        engineer_resp, engineer_explanation = generate_engineer_response(
-            user_text=user_text,
-            tokenizer=tokenizerE,
-            model=modelE
-        )
-        st.session_state.conversation.append(("Engineer", engineer_resp))
-        st.session_state.conversation.append(("Engineer Explanation", engineer_explanation))
-
-        # Display Engineer response immediately
-        st.markdown(f"**Engineer:** {engineer_resp}")
-        st.markdown(f"<i>{engineer_explanation}</i>", unsafe_allow_html=True)
-
-        # Analyst generates a response based on engineer's output
-        analyst_resp, analyst_explanation = generate_analyst_response(
-            user_text=user_text,
-            engineer_output=engineer_resp,
-            tokenizer=tokenizerA,
-            model=modelA
-        )
-        st.session_state.conversation.append(("Analyst", analyst_resp))
-        st.session_state.conversation.append(("Analyst Explanation", analyst_explanation))
-
-        # Display Analyst response immediately
-        st.markdown(f"**Analyst:** {analyst_resp}")
-        st.markdown(f"<i>{analyst_explanation}</i>", unsafe_allow_html=True)
-
-        # Limit the conversation to 3 exchanges between Engineer and Analyst
-        for _ in range(2):
+        with st.spinner("Engineer is formulating a solution..."):
             engineer_resp, engineer_explanation = generate_engineer_response(
-                user_text=analyst_resp,
+                user_text=user_text,
                 tokenizer=tokenizerE,
                 model=modelE
             )
             st.session_state.conversation.append(("Engineer", engineer_resp))
             st.session_state.conversation.append(("Engineer Explanation", engineer_explanation))
 
-            # Display Engineer response immediately
-            st.markdown(f"**Engineer:** {engineer_resp}")
-            st.markdown(f"<i>{engineer_explanation}</i>", unsafe_allow_html=True)
+        # Display Engineer response immediately
+        st.markdown(f"**Engineer:** {engineer_resp}")
+        st.markdown(f"<i>{engineer_explanation}</i>", unsafe_allow_html=True)
 
+        # Analyst generates a response based on engineer's output
+        with st.spinner("Analyst is analyzing data and providing insights..."):
             analyst_resp, analyst_explanation = generate_analyst_response(
-                user_text=engineer_resp,
+                user_text=user_text,
                 engineer_output=engineer_resp,
                 tokenizer=tokenizerA,
                 model=modelA
@@ -158,14 +133,43 @@ if st.button("Start/Continue Conversation"):
             st.session_state.conversation.append(("Analyst", analyst_resp))
             st.session_state.conversation.append(("Analyst Explanation", analyst_explanation))
 
-            # Display Analyst response immediately
-            st.markdown(f"**Analyst:** {analyst_resp}")
-            st.markdown(f"<i>{analyst_explanation}</i>", unsafe_allow_html=True)
+        # Display Analyst response immediately
+        st.markdown(f"**Analyst:** {analyst_resp}")
+        st.markdown(f"<i>{analyst_explanation}</i>", unsafe_allow_html=True)
+
+        # Limit the conversation to 3 exchanges between Engineer and Analyst
+        for _ in range(2):
+            with st.spinner("Engineer is formulating a solution..."):
+                engineer_resp, engineer_explanation = generate_engineer_response(
+                    user_text=analyst_resp,
+                    tokenizer=tokenizerE,
+                    model=modelE
+                )
+                st.session_state.conversation.append(("Engineer", engineer_resp))
+                st.session_state.conversation.append(("Engineer Explanation", engineer_explanation))
+
+                # Display Engineer response immediately
+                st.markdown(f"**Engineer:** {engineer_resp}")
+                st.markdown(f"<i>{engineer_explanation}</i>", unsafe_allow_html=True)
+
+            with st.spinner("Analyst is analyzing data and providing insights..."):
+                analyst_resp, analyst_explanation = generate_analyst_response(
+                    user_text=engineer_resp,
+                    engineer_output=engineer_resp,
+                    tokenizer=tokenizerA,
+                    model=modelA
+                )
+                st.session_state.conversation.append(("Analyst", analyst_resp))
+                st.session_state.conversation.append(("Analyst Explanation", analyst_explanation))
+
+                # Display Analyst response immediately
+                st.markdown(f"**Analyst:** {analyst_resp}")
+                st.markdown(f"<i>{analyst_explanation}</i>", unsafe_allow_html=True)
 
         # Generate the summary after the conversation
         final_plan = summarize_conversation(st.session_state.conversation)
         st.session_state.conversation.append(("Summary", final_plan))
-        st.markdown(f"**Summary:** {final_plan}")
+        st.markdown(final_plan)
 
 for speaker, text in st.session_state.conversation:
     if speaker == "User":
@@ -174,3 +178,5 @@ for speaker, text in st.session_state.conversation:
         st.markdown(f"**{speaker}:** {text}")
     elif "Explanation" in speaker:
         st.markdown(f"<i>{speaker}: {text}</i>", unsafe_allow_html=True)
+    else:
+        st.markdown(f"**{speaker}:** {text}")