app.py

# app.py - Flask Backend
from flask import Flask, request, jsonify, send_from_directory
import google.generativeai as genai
from dotenv import load_dotenv
import os
from flask_cors import CORS
import markdown2
import re
from gtts import gTTS # <-- Import gTTS
import uuid # <-- Import UUID for unique filenames

# Load environment variables
load_dotenv()

# Define paths and create static audio directory if it doesn't exist
AUDIO_FOLDER = os.path.join('static', 'audio')
if not os.path.exists(AUDIO_FOLDER):
    os.makedirs(AUDIO_FOLDER)

# Initialize Flask app
app = Flask(__name__, static_folder='static')
CORS(app)  # Enable CORS for all routes

# Configure Gemini with a system instruction
# This guides the AI's behavior and ensures responses are good for TTS.
system_instruction_text = """
You are a helpful, friendly, and informative AI assistant named AstroChat.
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.
"""

genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
# Initialize the model with the system instruction
model = genai.GenerativeModel(
    'gemini-2.5-flash', # Using 1.5-flash for better performance and system_instruction support
    system_instruction=system_instruction_text
)

def convert_markdown_to_html(text):
    # Convert markdown to HTML
    # Using 'fenced-code-blocks' and 'tables' for better markdown support
    html = markdown2.markdown(text, extras=["fenced-code-blocks", "tables"])
    
    # Add custom styling to code blocks (pre blocks)
    # This specifically targets `<pre><code>` blocks generated by markdown2 for styling.
    html = re.sub(r'<pre><code(.*?)>', r'<pre class="code-block"><code\1>', html)
    
    # Convert **bold** to <strong> for better visibility (markdown2 usually handles this, but good to ensure)
    html = re.sub(r'\*\*(.*?)\*\*', r'<strong>\1</strong>', html)
    
    # Convert *italic* to <em> (markdown2 usually handles this, but good to ensure)
    html = re.sub(r'\*(.*?)\*', r'<em>\1</em>', html)
    
    return html

@app.route('/chat', methods=['POST'])
def chat():
    try:
        data = request.json
        user_message = data.get('message')
        
        if not user_message:
            return jsonify({"error": "No message provided"}), 400
        
        # Generate response using Gemini
        # For multi-turn conversations, you might manage chat history here
        response = model.generate_content(user_message)
        
        # Get plain text for audio generation
        plain_text_response = response.text
        
        # Convert markdown to HTML for display
        html_response = convert_markdown_to_html(plain_text_response)
        
        return jsonify({
            "response_html": html_response,
            "response_text": plain_text_response # Send plain text for TTS
        })
        
    except Exception as e:
        app.logger.error(f"Chat Error: {e}")
        return jsonify({"error": str(e)}), 500

@app.route('/generate-audio', methods=['POST'])
def generate_audio():
    try:
        data = request.json
        text_to_speak = data.get('text')

        if not text_to_speak:
            return jsonify({"error": "No text provided"}), 400

        # Sanitize text for TTS (remove common markdown characters for smoother pronunciation)
        # This prevents gTTS from trying to pronounce asterisks, backticks, etc.
        cleaned_text = re.sub(r'[\*_`#]', '', text_to_speak) # Remove bold, italic, code, headers markdown
        cleaned_text = re.sub(r'\s+', ' ', cleaned_text).strip() # Replace multiple spaces with single space

        if not cleaned_text: # If text becomes empty after cleaning
            return jsonify({"error": "Text became empty after cleaning, cannot generate audio."}), 400

        # Generate a unique filename using UUID to prevent collisions
        filename = f"{uuid.uuid4()}.mp3"
        filepath = os.path.join(AUDIO_FOLDER, filename)

        # Create TTS object and save to file
        tts = gTTS(text=cleaned_text, lang='en', slow=False) # 'en' for English, 'slow=False' for normal speed
        tts.save(filepath)

        # Return the URL to the audio file, converting path separators for web use
        audio_url = f"/{filepath.replace(os.path.sep, '/')}"
        return jsonify({"audio_url": audio_url})

    except Exception as e:
        app.logger.error(f"Audio Generation Error: {e}")
        return jsonify({"error": str(e)}), 500

# Serve the main index.html file
@app.route('/')
def serve_index():
    return send_from_directory('static', 'index.html')

# Serve other static files (CSS, JS, audio files)
@app.route('/<path:path>')
def serve_static(path):
    # Ensure that only files from 'static' are served
    return send_from_directory('static', path)

if __name__ == '__main__':
    # Run the Flask app
    # debug=True allows automatic reloading on code changes and provides more detailed error messages
    app.run(host="0.0.0.0", port=7860)