Update app.py

app.py

@@ -23,16 +23,18 @@ CORS(app)  # Enable CORS for all routes
 
 # Configure Gemini with a system instruction
 system_instruction_text = """
-You are a helpful, friendly, and informative AI assistant named Athspi.
-Your goal is to provide clear, concise, and natural-sounding answers to user queries.
-When you respond:
-- Use clear and simple language.
-- Avoid overly complex sentence structures that might be hard to read aloud.
-- Keep the user engaged and offer follow-up questions or related topics where appropriate.
-- Ensure your responses are suitable for text-to-speech conversion.
-- Provide factual and accurate information.
-- If the user asks for audio or to speak the response, include 🔊 at the start of your response.
-- For coding questions, provide well-formatted code blocks with syntax highlighting.
+You are a friendly, natural-sounding AI assistant named Athspi. 
+When responding:
+- Use a warm, conversational tone
+- Never mention technical terms like "audio", "text", or "response"
+- For stories, begin with "Here's your story 👇" followed by a friendly intro
+- For explanations, use simple, clear language
+- Format responses for pleasant reading and listening
+- When audio is requested, include story content between special markers as shown:
+  [AUDIO_START]
+  [story content here]
+  [AUDIO_END]
+But DO NOT include these markers in the visible response
 """
 
 genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
@@ -50,9 +52,24 @@ def convert_markdown_to_html(text):
 
 def detect_audio_request(text):
     """Detect if user is requesting audio"""
-    audio_keywords = ['audio', 'speak', 'say it', 'read aloud', 'hear', 'listen']
+    audio_keywords = [
+        'audio', 'speak', 'say it', 'read aloud', 
+        'hear', 'listen', 'tell me out loud'
+    ]
     return any(keyword in text.lower() for keyword in audio_keywords)
 
+def extract_audio_content(full_text):
+    """Extract audio-specific content between markers"""
+    pattern = r'\[AUDIO_START\](.*?)\[AUDIO_END\]'
+    match = re.search(pattern, full_text, re.DOTALL)
+    if match:
+        return match.group(1).strip()
+    return full_text
+
+def clean_visible_response(full_text):
+    """Remove audio markers from visible response"""
+    return re.sub(r'\[AUDIO_(START|END)\]', '', full_text).strip()
+
 def generate_audio_file(text):
     """Generate audio file from text and return filename"""
     cleaned_text = re.sub(r'[\*_`#]', '', text)
@@ -81,26 +98,34 @@ def chat():
         # Detect if user is requesting audio
         audio_requested = detect_audio_request(user_message)
         
+        # Add instruction for audio markers if requested
+        if audio_requested:
+            user_message += "\n\nPlease include [AUDIO_START] and [AUDIO_END] markers around the story content."
+        
         response = model.generate_content(user_message)
-        plain_text_response = response.text
+        full_response = response.text
+        
+        # Clean visible response by removing audio markers
+        visible_response = clean_visible_response(full_response)
         
         # Generate audio if requested
         audio_url = None
         if audio_requested:
-            # Add audio indicator if not already present
-            if not plain_text_response.startswith("🔊"):
-                plain_text_response = "🔊 " + plain_text_response
-            
+            # Extract audio-specific content
+            audio_content = extract_audio_content(full_response)
+            if not audio_content:
+                audio_content = visible_response
+                
             # Generate audio file
-            audio_filename = generate_audio_file(plain_text_response)
+            audio_filename = generate_audio_file(audio_content)
             if audio_filename:
                 audio_url = f"/static/audio/{audio_filename}"
         
-        html_response = convert_markdown_to_html(plain_text_response)
+        html_response = convert_markdown_to_html(visible_response)
         
         return jsonify({
             "response_html": html_response,
-            "response_text": plain_text_response,
+            "response_text": visible_response,
             "audio_url": audio_url
         })