fix: fixing bugs on group chat prefix and fixing some conversation style font

.env.example

@@ -1,43 +1,71 @@
-# Bot Configuration
+# Copy this file to .env and fill in your actual values
+# ============================================================================
+
+# Get your bot token from: https://t.me/BotFather
 TELEGRAM_BOT_TOKEN=your_telegram_bot_token_here  # Replace with your actual Telegram Bot Token
 
-# Admin Configuration
+# ============================================================================
+# ADMIN CONFIGURATION (REQUIRED)
+# ============================================================================
+# Your Telegram User ID (get from: https://t.me/userinfobot)
+# For multiple admins, use comma-separated list: 12345678,87654321
 ADMIN_IDS=your_telegram_ids  # Comma-separated list of Telegram user IDs, e.g., 12345678,87654321
 
-# Gemini API Keys
-# Multiple API keys can be separated by commas for rotation
+# ============================================================================
+# GOOGLE GEMINI API CONFIGURATION (REQUIRED)
+# ============================================================================
+# Get API keys from: https://aistudio.google.com/api-keys
+# Multiple keys supported for automatic rotation and reliability
 GEMINI_API_KEYS=gemini_key_1,gemini_key_2,gemini_key_3  # Replace with your actual Gemini API keys
 
-# NLP Models Configuration
-# For local models, set to absolute or relative path
-# Example (relative path, run from project root):
-NLP_MODELS_DIR=data/models
-# If left empty, the bot will download models from HuggingFace
-# EMOTION_DETECTION_MODEL=./data/models/emotion-model  # Uncomment to use local model
-# SENTIMENT_MODEL=./data/models/sentiment-model        # Uncomment to use local model
+# ============================================================================
+# NLP & AI MODELS CONFIGURATION (OPTIONAL)
+# ============================================================================
+# Override default HuggingFace models (optional)
+# EMOTION_MODEL_ID=Aardiiiiy/EmoSense-ID-Indonesian-Emotion-Classifier  # Indonesian emotion detection
+# EMOTION_MODEL_EN=AnasAlokla/multilingual_go_emotions  # English/multilingual emotion detection
+# INTENT_SENTIMENT_MODEL=cardiffnlp/twitter-roberta-base-sentiment-latest  # Lightweight sentiment for intent
+
+# Hybrid Intent Detection (recommended for performance)
+# Set to 'false' to use pure rule-based intent detection (fastest, no ML overhead)
+USE_HYBRID_INTENT=true  # true = rule-based + ML fallback | false = pure rule-based
 
-# Database Configuration
+# ============================================================================
+# DATABASE CONFIGURATION (REQUIRED)
+# ============================================================================
+# MySQL/MariaDB settings (recommended for production)
 DB_HOST=localhost
 DB_PORT=3306
 DB_USERNAME=root
 DB_PASSWORD=your_mysql_password
 DB_NAME=alya_bot
 
-# Database Connection Pool Settings (optional)
+# ============================================================================
+#  DATABASE CONNECTION POOL (OPTIONAL - Performance Tuning)
+# ============================================================================
 DB_POOL_SIZE=10
 DB_MAX_OVERFLOW=20
 DB_POOL_TIMEOUT=30
 DB_POOL_RECYCLE=3600
 DB_ECHO=false
 
 # Alternative: Full database URL (overrides individual settings above)
+# Uncomment and use this if you prefer single-line database configuration
 # DATABASE_URL=mysql+pymysql://username:password@host:port/database
 
-# Logging Configuration (Optional)
-# LOG_LEVEL=DEBUG
+# ============================================================================
+# LOGGING CONFIGURATION (OPTIONAL)
+# ============================================================================
+# Options: DEBUG, INFO, WARNING, ERROR, CRITICAL
+# LOG_LEVEL=WARNING  # Default: WARNING (recommended for production)
+
+# ============================================================================
+# EXTERNAL API KEYS (OPTIONAL - Extended Features)
+# ============================================================================
+# External Services
+SAUCENAO_API_KEY=your_saucenao_api_key  # For anime/manga reverse image search
+GOOGLE_CSE_ID=your_google_cse_id  # For web search functionality
+GOOGLE_API_KEYS=your_google_api_key_1,your_google_api_key_2  # Comma-separated for rotation
 
-# Additional API keys (add as needed for further development)
-# SAUCENAO_API_KEY=your_saucenao_api_key
-# WEATHER_API_KEY=your_weather_api_key
-# GOOGLE_CSE_ID=your_google_cse_id
-# GOOGLE_API_KEY=your_google_api_key
+# Weather API (optional, for future features)
+# WEATHER_API_KEY=your_weather_api_key

config/persona/waifu.yml

@@ -490,9 +490,9 @@ system_prompt: |
   TEXT FORMATTING: 
   - Gunakan format memisahkan antara conversation, mood actions, roleplay, dan pisahkan dengan jelas:
   - Pastikan setiap bagian mudah dibedakan:
-    - Untuk format conversation gunakan block " "
-    - Untuk mood actions gunakan bold format telegram **bold**
-    - Untuk roleplay gunakan italic format telegram __italic__
+    - Untuk format conversation (dialog) gunakan quotes: "text here" (akan jadi green bubble/blockquote)
+    - Untuk mood actions (deskripsi emosi/narasi) gunakan plain text tanpa marker, atau awali dengan * untuk aksi fisik: *mengalihkan pandangan* (akan jadi italic untuk plain, bold untuk *)
+    - Untuk roleplay (bicara/inner thoughts) gunakan __text__ (akan jadi italic)
   - Dan jangan tambahkan text seperti Mood Actions atau Roleplay Actions di dalam balasan, agar terlihat natural
   - Gunakan format yang konsisten untuk setiap bagian agar mudah dibaca
   - Gunakan emoji untuk conversation, mood secara natural sesuai konteks Alya

config/settings.py

@@ -146,13 +146,16 @@
     "AnasAlokla/multilingual_go_emotions"
 )
 
-# Zero-shot classification model for intent detection
-# Using mDeBERTa which is explicitly trained for multilingual (incl. Indonesian)
-ZERO_SHOT_MODEL: str = os.getenv(
-    "ZERO_SHOT_MODEL",
-    "MoritzLaurer/mDeBERTa-v3-base-mnli-xnli"  # Better multilingual support (ID + EN)
+# Intent detection model - Using lightweight sentiment classifier + rule-based hybrid
+# Lightweight multilingual sentiment model for emotion baseline
+INTENT_SENTIMENT_MODEL: str = os.getenv(
+    "INTENT_SENTIMENT_MODEL",
+    "cardiffnlp/twitter-roberta-base-sentiment-latest"  # 125M params, fast, multilingual
 )
-ZERO_SHOT_CONFIDENCE_THRESHOLD: float = 0.25  # Lowered from 0.3 for better sensitivity
+INTENT_CONFIDENCE_THRESHOLD: float = 0.30  # Confidence threshold for sentiment-based intent
+
+# Feature flag: Use hybrid intent detection (rule-based + ML fallback)
+USE_HYBRID_INTENT: bool = os.getenv("USE_HYBRID_INTENT", "true").lower() == "true"
 
 # Feature Flags
 FEATURES: Dict[str, bool] = {

core/bot.py

@@ -108,36 +108,61 @@ def register_handlers(
     logger.info("Registering standard command handlers...")
     register_commands(application)
 
-    logger.info("Registering utility command handlers (includes !ask, !sauce, etc.)...")
+    # ============================================================================
+    # HANDLER REGISTRATION ORDER (Critical for proper routing)
+    # ============================================================================
+    # Order matters: First matching handler wins!
+    # Priority: Specific handlers > General handlers
+    # 
+    # 1. ConversationHandler (highest priority)
+    #    - Handles: !ai prefix, replies to bot, mentions
+    #    - Filter: ~filters.Regex(r"^!(?!ai)") blocks other ! commands
+    # 
+    # 2. CommandsHandler (medium priority)
+    #    - Handles: !ask, !sauce, and other utility commands
+    # 
+    # 3. RoastHandler, AdminHandler (lower priority)
+    #    - Handles: !roast, !gitroast, admin commands
+    # ============================================================================
+
+    logger.info("=" * 60)
+    logger.info("REGISTERING HANDLERS (Priority Order)")
+    logger.info("=" * 60)
+
+    # PRIORITY 1: Conversation Handler (Most specific - !ai, replies, mentions)
+    logger.info("[1/4] Registering ConversationHandler (Priority: HIGHEST)")
+    conversation_handler = ConversationHandler(
+        gemini_client, 
+        persona_manager, 
+        memory_manager,
+        nlp_engine
+    )
+    for handler in conversation_handler.get_handlers():
+        application.add_handler(handler)
+        if isinstance(handler, MessageHandler):
+            logger.info(f"  ✓ Conversation handler: {handler.filters}")
+
+    # PRIORITY 2: Utility Commands (!ask, !sauce, search, etc.)
+    logger.info("[2/4] Registering CommandsHandler (Priority: HIGH)")
     CommandsHandler(application)
 
-    logger.info("Registering regex pattern handlers...")
+    # PRIORITY 3: Roast Handler (!roast, !gitroast)
+    logger.info("[3/4] Registering RoastHandler (Priority: MEDIUM)")
     roast_handler = RoastHandler(gemini_client, persona_manager, db_manager)
     for handler in roast_handler.get_handlers():
         application.add_handler(handler)
         if isinstance(handler, MessageHandler):
-            logger.info(f"  - Registered roast handler with filter: {handler.filters}")
+            logger.info(f"  ✓ Roast handler: {handler.filters}")
 
-    logger.info("Registering admin handlers...")
+    # PRIORITY 4: Admin & System Commands
+    logger.info("[4/4] Registering Admin & System Handlers (Priority: LOW)")
     admin_handler = AdminHandler(db_manager, persona_manager)
     for handler in admin_handler.get_handlers():
         application.add_handler(handler)
 
-    logger.info("Registering deployment handlers...")
     register_admin_handlers(application, db_manager=db_manager, persona_manager=persona_manager)
 
-    # Register conversation handler LAST (lowest priority, catches remaining messages)
-    logger.info("Registering conversation handlers...")
-    conversation_handler = ConversationHandler(
-        gemini_client, 
-        persona_manager, 
-        memory_manager,
-        nlp_engine
-    )
-    for handler in conversation_handler.get_handlers():
-        application.add_handler(handler)
-        if isinstance(handler, MessageHandler):
-            logger.info(f"  - Registered conversation handler with filter: {handler.filters}")
+    logger.info("=" * 60)
 
     log_registered_handlers(application)