Update app.py

app.py

@@ -2,60 +2,75 @@ import gradio as gr
 import json
 import asyncio
 import edge_tts
+import re
 import os
 from huggingface_hub import InferenceClient
 
-# --- CONFIGURATION ---
-# We use Llama-3 for BOTH tasks now as it is the most reliable on the free tier
+# --- SETTINGS ---
+# We use the Llama-3 model for everything because it is smart and free.
 EXTRACTOR_MODEL = "meta-llama/Meta-Llama-3-8B-Instruct"
 PERSONALITY_MODEL = "meta-llama/Meta-Llama-3-8B-Instruct"
 
-# Default Data
+# Default Chat History (30 messages showing different personality traits)
 DEFAULT_LOGS = """
-1. User: I feel completely drained after large social gatherings.
-2. User: I enjoy abstract theories more than practical details.
-3. User: My desk is always messy, but I know where everything is.
-4. User: I often worry that I said the wrong thing in conversations.
-5. User: I prefer planning everything weeks in advance. Spontaneity stresses me out.
-6. User: I find it hard to sympathize with people who are overly emotional.
-7. User: I love exploring new ideas and hobbies, even if I don't finish them.
-8. User: Criticism hits me really hard, even if it's constructive.
-9. User: I tend to take charge in group projects because I want things done right.
-10. User: I value logic and truth over sparing someone's feelings.
-11. User: I often get lost in my own thoughts and forget my surroundings.
-12. User: I hate conflict. I will do anything to keep the peace.
-13. User: I feel a strong need to help others, sometimes at my own expense.
-14. User: I get bored easily with routine tasks.
-15. User: I am very skeptical until I see concrete evidence.
-16. User: I love meeting new people and being the center of attention.
-17. User: I struggle to express my emotions verbally.
-18. User: I often procrastinate until the very last minute.
-19. User: I feel deeply moved by art and music.
-20. User: I prefer a few close friends over a large group of acquaintances.
-21. User: I have a hard time saying "no" to people.
-22. User: I am constantly analyzing why people do what they do.
-23. User: I value tradition and established rules.
-24. User: I am often described as "intense" or "serious."
-25. User: I feel energetic when I'm debating intellectual topics.
-26. User: I worry a lot about the future.
-27. User: I trust my gut instincts more than data sometimes.
-28. User: I prefer to work alone rather than in a team.
-29. User: I am very competitive and hate losing.
-30. User: I want to understand my true purpose in life.
+1. User: I feel tired after big parties. I need to be alone to recharge.
+2. User: I like ideas more than real-world details.
+3. User: My desk is messy, but I know where my stuff is.
+4. User: I worry that I said the wrong thing.
+5. User: I like to plan ahead. Surprises stress me out.
+6. User: It is hard for me to understand why people cry over small things.
+7. User: I start many hobbies but do not finish them.
+8. User: I feel bad when someone criticizes me.
+9. User: I take charge in groups to make sure work is done right.
+10. User: Logic is more important than feelings.
+11. User: I daydream a lot.
+12. User: I hate fighting. I want everyone to get along.
+13. User: I help others even if it hurts me.
+14. User: Boring tasks make me sleepy.
+15. User: I need proof before I believe something.
+16. User: I love being the center of attention.
+17. User: I am bad at talking about my feelings.
+18. User: I wait until the last minute to do work.
+19. User: Music makes me feel strong emotions.
+20. User: I prefer 2 close friends over 20 acquaintances.
+21. User: I cannot say "no" to people.
+22. User: I always analyze why people act the way they do.
+23. User: I like following rules and traditions.
+24. User: People say I am too serious.
+25. User: I have lots of energy when debating.
+26. User: I am scared of the future.
+27. User: I trust my gut feeling more than numbers.
+28. User: I work better alone.
+29. User: I hate losing games.
+30. User: I want to know my purpose in life.
 """
 
-# --- LOGIC: MEMORY EXTRACTION ---
+# --- HELPER FUNCTION ---
+def clean_text_for_audio(text):
+    """
+    This removes text in brackets like (sighs) or *laughs*.
+    We do this so the voice robot does not read them out loud.
+    """
+    clean = re.sub(r'[\(\[\*].*?[\)\]\*]', '', text)
+    return clean.strip()
+
+# --- PART 1: MEMORY EXTRACTOR ---
 def extract_memory(chat_logs, hf_token):
     if not hf_token:
-        return json.dumps({"error": "Missing HF Token"}, indent=2)
+        return "Error: Please paste your Hugging Face Token."
     
     client = InferenceClient(token=hf_token)
     
+    # Instructions for the AI to find personality traits
     system_prompt = """
-    You are a Psychological Analyst. Analyze the chat logs.
-    Extract a profile with these JSON KEYS: 
-    "traits" (Big Five/MBTI), "values" (Logic, Harmony, etc), "struggles" (Anxiety, Procrastination).
-    OUTPUT: Valid JSON only. Do not add markdown blocks like ```json.
+    Read the chat logs. Create a simple User Profile in JSON format.
+    
+    Find these 3 things:
+    1. "traits": Is the user Introverted? Organized? Anxious?
+    2. "values": Do they care about Logic? Peace? Winning?
+    3. "struggles": Do they procrastinate? Have social anxiety?
+    
+    Return ONLY valid JSON.
     """
     
     messages = [
@@ -67,48 +82,47 @@ def extract_memory(chat_logs, hf_token):
         response = client.chat_completion(
             model=EXTRACTOR_MODEL,
             messages=messages,
-            max_tokens=600,
+            max_tokens=500,
             temperature=0.1
         )
         
-        # Parse JSON
+        # Clean up the answer to get just the JSON data
         text = response.choices[0].message.content.strip()
-        
-        # Cleaning potential markdown formatting from LLM response
-        if "```json" in text:
+        if "```" in text:
             text = text.replace("```json", "").replace("```", "")
-        
+            
         start = text.find("{")
         end = text.rfind("}") + 1
         return json.dumps(json.loads(text[start:end]), indent=2)
     except Exception as e:
         return json.dumps({"error": str(e)}, indent=2)
 
-# --- LOGIC: PERSONALITY RESPONSE ---
+# --- PART 2: PERSONALITY & VOICE ---
 async def generate_response_and_audio(message, memory_json, persona, hf_token):
     if not hf_token:
-        return "Please enter HF Token.", None
+        return "Error: Please paste your Hugging Face Token.", None
         
     client = InferenceClient(token=hf_token)
     
+    # Load the user's memory profile
     try:
         memory = json.loads(memory_json)
     except:
         memory = {}
 
-    # 1. Define Persona Prompts
+    # Define the 3 Personalities
     prompts = {
-        "Calm Mentor": "Role: Mentor. Tone: Calm, wise, patient. Focus: Growth, long-term perspective. Voice: Deep & Slow.",
-        "Witty Friend": "Role: Friend. Tone: Witty, casual, sarcastic, fun. Focus: Relatability, humor. Voice: Fast & Energetic.",
-        "Therapist-style": "Role: Therapist. Tone: Empathetic, soft, validating. Focus: Emotions, safety. Voice: Soft & Warm."
+        "Calm Mentor": "Role: Wise Teacher. Tone: Calm, slow, patient. Advice: Focus on long-term growth.",
+        "Witty Friend": "Role: Best Friend. Tone: Funny, fast, sarcastic. Advice: Make jokes and be relatable.",
+        "Therapist": "Role: Counselor. Tone: Soft, kind, gentle. Advice: Validate their feelings and ask how they feel."
     }
     
-    # 2. Context Injection
+    # Add User Memory to the AI's brain
     context = f"""
-    USER PROFILE:
-    - Traits: {memory.get('traits', 'Unknown')}
+    ABOUT THE USER:
+    - Personality: {memory.get('traits', 'Unknown')}
     - Values: {memory.get('values', 'Unknown')}
-    - Struggles: {memory.get('struggles', 'Unknown')}
+    - Problems: {memory.get('struggles', 'Unknown')}
     """
     
     messages = [
@@ -117,7 +131,7 @@ async def generate_response_and_audio(message, memory_json, persona, hf_token):
     ]
     
     try:
-        # 3. Generate Text
+        # Step A: Write the text response
         res = client.chat_completion(
             model=PERSONALITY_MODEL,
             messages=messages,
@@ -126,57 +140,60 @@ async def generate_response_and_audio(message, memory_json, persona, hf_token):
         )
         text_reply = res.choices[0].message.content
         
-        # 4. Generate Audio (Edge-TTS)
+        # Step B: Clean text for the voice engine
+        spoken_text = clean_text_for_audio(text_reply)
+        
+        # Step C: Pick a voice based on the persona
         voice_map = {
-            "Calm Mentor": "en-US-ChristopherNeural",
-            "Witty Friend": "en-US-EricNeural",
-            "Therapist-style": "en-US-AvaNeural"
+            "Calm Mentor": "en-US-ChristopherNeural", # Deep Male Voice
+            "Witty Friend": "en-US-EricNeural",       # Fast Male Voice
+            "Therapist": "en-US-AvaNeural"            # Soft Female Voice
         }
         
+        # Step D: Create the audio file
         output_file = "response.mp3"
-        communicate = edge_tts.Communicate(text_reply, voice_map.get(persona, "en-US-AriaNeural"))
+        communicate = edge_tts.Communicate(spoken_text, voice_map.get(persona, "en-US-AriaNeural"))
         await communicate.save(output_file)
         
         return text_reply, output_file
         
     except Exception as e:
-        return f"API Error: {str(e)}", None
+        return f"Error: {str(e)}", None
 
-# Wrapper for Gradio
+# Wrapper to make the async code work with Gradio
 def process_interaction(message, memory_json, persona, hf_token):
     return asyncio.run(generate_response_and_audio(message, memory_json, persona, hf_token))
 
-# --- GRADIO UI ---
+# --- PART 3: THE USER INTERFACE ---
 with gr.Blocks(title="Personality Engine") as demo:
-    gr.Markdown("# Personality Engine: Memory & Tones")
+    gr.Markdown("See how Memory + Tone changes the AI response.")
     
     with gr.Row():
-        token_input = gr.Textbox(label="Hugging Face Token", type="password", placeholder="Paste HF Token here...")
+        token_input = gr.Textbox(label="Hugging Face Token (Required)", type="password")
     
     with gr.Row():
-        # Left Column: Memory
-        with gr.Column(scale=1):
-            gr.Markdown("### 1. Memory Extraction")
-            logs_input = gr.Textbox(label="Chat Logs (Input)", value=DEFAULT_LOGS, lines=10)
-            extract_btn = gr.Button("Analyze Profile")
-            memory_output = gr.Code(label="Extracted Memory (JSON)", language="json")
+        # Column 1: Analyze the User
+        with gr.Column():
+            gr.Markdown("### 1. Analyze User History")
+            logs_input = gr.Textbox(label="Chat History", value=DEFAULT_LOGS, lines=10)
+            extract_btn = gr.Button("Create Profile")
+            memory_output = gr.Code(label="Result (JSON)", language="json")
             
             extract_btn.click(extract_memory, inputs=[logs_input, token_input], outputs=memory_output)
 
-        # Right Column: Interaction
-        with gr.Column(scale=1):
-            gr.Markdown("### 2. Personality Interaction")
-            user_msg = gr.Textbox(label="Your Message", value="I'm feeling overwhelmed by the future.")
+        # Column 2: Chat with Persona
+        with gr.Column():
+            gr.Markdown("### 2. Chat with Agent")
+            user_msg = gr.Textbox(label="Your Message", value="I am scared about the future.")
             persona_select = gr.Radio(
-                ["Calm Mentor", "Witty Friend", "Therapist-style"], 
-                label="Select Tone", 
+                ["Calm Mentor", "Witty Friend", "Therapist"], 
+                label="Choose Personality", 
                 value="Calm Mentor"
             )
-            generate_btn = gr.Button("Generate Response")
+            generate_btn = gr.Button("Get Answer")
             
-            with gr.Group():
-                text_output = gr.Markdown(label="Response Text")
-                audio_output = gr.Audio(label="Voice Response")
+            text_output = gr.Markdown(label="Text Answer")
+            audio_output = gr.Audio(label="Voice Answer")
             
             generate_btn.click(
                 process_interaction,