Update app.py

app.py

@@ -2,8 +2,7 @@ import asyncio
 import streamlit as st
 import os
 import pandas as pd
-from together import Together
-from helper import *
+from helper import call_llama  # Make sure this is properly defined in helper.py
 
 st.set_page_config(layout="wide")
 st.title("Duel Agent Simulation 🦙🦙")
@@ -12,24 +11,16 @@ with st.sidebar:
     with st.expander("Instruction Manual"):
         st.markdown("""
             # 🦙🦙 Duel Agent Simulation Streamlit App
-
             ## Overview
-
             Welcome to the **Duel Agent Simulation** app! This Streamlit application allows you to chat with Meta's Llama3 model in a unique interview simulation. The app features two agents in an interview scenario, with a judge providing feedback. The best part? You simply provide a topic, and the simulation runs itself!
-
             ## Features
-
             ### 📝 Instruction Manual
-
             **Meta Llama3 🦙 Chatbot**
-
             This application lets you interact with Meta's Llama3 model through a fun interview-style chat.
-
             **How to Use:**
             1. **Input:** Type a topic into the input box labeled "Enter a topic".
-            2. **Submit:** Press the "Submit" button to start the simulation.
+            2. **Submit:** Press the "Run Simulation!" button to start the simulation.
             3. **Chat History:** View the previous conversations as the simulation unfolds.
-
             **Credits:**
             - **Developer:** Kethan Dosapati 
                - [LinkedIn](https://www.linkedin.com/in/kethan-dosapati/)  
@@ -44,11 +35,22 @@ with st.sidebar:
     # Add a button to clear the session state
     if st.button("Clear Session"):
         st.session_state.messages = []
+        st.session_state.simulation_data = []
         st.experimental_rerun()
 
 # Initialize chat history
 if "messages" not in st.session_state:
     st.session_state.messages = []
+    
+# Initialize simulation data storage
+if "simulation_data" not in st.session_state:
+    st.session_state.simulation_data = {
+        "iterations": [],
+        "questions": [],
+        "answers": [],
+        "judge_comments": [],
+        "passes": []
+    }
 
 # Display chat messages from history on app rerun
 for message in st.session_state.messages:
@@ -61,82 +63,68 @@ interviewee = call_llama
 judge = call_llama
 
 # React to user input
-iter = 0
-list_of_iters = []
-list_of_questions = []
-list_of_answers = []
-list_of_judge_comments = []
-list_of_passes = []
 if submit_button:
-
-    # Initiatization
+    iter = 0
     prompt = f"Ask a question about this topic: {user_topic}"
 
     # Display user message in chat message container
-    # Default: user=interviewee, assistant=interviewer
-    st.chat_message("user").markdown(prompt)
-    st.session_state.messages.append({"role": "user", "content": prompt})
-
-    while True:
-
-        # Interview asks a question
-        question = interviewer(prompt)
-
-        # Display assistant response in chat message container
-        st.chat_message("assistant").markdown(question)
-        st.session_state.messages.append({"role": "assistant", "content": question})
-
-        # Interviewee attempts an answer
-        if iter < 5:
-            answer = interviewee(
-                f"""
-                    Answer the question: {question} in a mediocre way
-                    Because you are an inexperienced interviewee.
-                """
-            )
-            st.chat_message("user").markdown(answer)
-            st.session_state.messages.append({"role": "user", "content": answer})
-        else:
-            answer = interviewee(
-                f"""
-                    Answer the question: {question} in a mediocre way
-                    Because you are an inexperienced interviewee but you really want to learn,
-                    so you learn from the judge comments: {judge_comments}
-                """
-            )
+    st.chat_message("user").markdown(f"Topic: {user_topic}")
+    st.session_state.messages.append({"role": "user", "content": f"Topic: {user_topic}"})
+
+    with st.spinner("Running simulation..."):
+        while True:
+            # Interviewer asks a question
+            question = interviewer(prompt)
+            
+            # Display question
+            st.chat_message("assistant").markdown(question)
+            st.session_state.messages.append({"role": "assistant", "content": question})
+
+            # Interviewee answers
+            if iter < 2:  # First few answers are mediocre
+                answer_prompt = f"Answer the question: {question} in a mediocre way because you are an inexperienced interviewee."
+            else:  # Later answers try to improve
+                judge_comments = st.session_state.simulation_data["judge_comments"][-1] if iter > 0 else ""
+                answer_prompt = f"""Answer the question: {question} in a mediocre way because you are an inexperienced interviewee, 
+                                but you're learning from these judge comments: {judge_comments}"""
+            
+            answer = interviewee(answer_prompt)
             st.chat_message("user").markdown(answer)
             st.session_state.messages.append({"role": "user", "content": answer})
 
-        # Judge thinks and advises but the thoughts are hidden
-        judge_comments = judge(
-            f"""
-                The question is: {question}
-                The answer is: {answer}
-                Provide feedback and rate the answer from 1 to 10 while 10 being the best and 1 is the worst. 
-            """
-        )
-
-        # Collect all responses
-        passed_or_not = 1 if '8' in judge_comments else 0
-        list_of_iters.append(iter)
-        list_of_questions.append(question)
-        list_of_answers.append(answer)
-        list_of_judge_comments.append(judge_comments)
-        list_of_passes.append(passed_or_not)
-        results_tab = pd.DataFrame({
-            "Iter.": list_of_iters,
-            "Questions": list_of_questions,
-            "Answers": list_of_answers,
-            "Judge Comments": list_of_judge_comments,
-            "Passed": list_of_passes
-        })
-
-        with st.expander("See explanation"):
-            st.table(results_tab)
-
-        # Stopping rule
-        if '8' in judge_comments:
-            break
-
-        # Checkpoint
-        iter += 1
+            # Judge evaluates
+            judge_prompt = f"""The question was: {question}
+                            The answer was: {answer}
+                            Provide brief feedback and rate the answer from 1 to 10 (10 being best)."""
+            judge_comments = judge(judge_prompt)
+            
+            # Store results
+            passed = 1 if any(str(i) in judge_comments for i in [8, 9, 10]) else 0
+            st.session_state.simulation_data["iterations"].append(iter)
+            st.session_state.simulation_data["questions"].append(question)
+            st.session_state.simulation_data["answers"].append(answer)
+            st.session_state.simulation_data["judge_comments"].append(judge_comments)
+            st.session_state.simulation_data["passes"].append(passed)
+            
+            # Show results table
+            with st.expander("Simulation Details"):
+                results_df = pd.DataFrame({
+                    "Iteration": st.session_state.simulation_data["iterations"],
+                    "Question": st.session_state.simulation_data["questions"],
+                    "Answer": st.session_state.simulation_data["answers"],
+                    "Judge Feedback": st.session_state.simulation_data["judge_comments"],
+                    "Passed?": ["✅" if x else "❌" for x in st.session_state.simulation_data["passes"]]
+                })
+                st.table(results_df)
+
+            # Stopping condition
+            if passed:
+                st.success("Interview passed! The candidate gave a good enough answer.")
+                break
+                
+            if iter >= 5:  # Max iterations
+                st.warning("Interview ended - maximum attempts reached")
+                break
+                
+            iter += 1
+            prompt = f"Ask another follow-up question about: {user_topic}"