HappyBot: A Multimodal Generative AI Telegram Assistant

HappyBot is a Telegram bot designed to provide emotional support, wellness check-ins, and intelligent conversations using a Generative AI backend (SEA-LION LLM). The bot integrates a Mini App (Telegram Web App), handles both text and voice input, and is deployed using webhook mode on Render.com.

📌 Features

🤖 LLM-powered responses (via SEA-LION API)
🧠 LangChain context-aware prompt formatting
🎙️ Voice message transcription and voice reply (Sphinx + gTTS)
📆 Weekly wellbeing check-in with Telegram polls
📲 Telegram Mini App for setting custom alerts (HTML/CSS/JS)
☁️ Hosted on Render using webhook deployment
🔁 Redis (or fallback in-memory) for context and conversation memory

⚙️ Tech Stack

Component	Tech Used
Backend	Python, Flask (for webhook), Telegram Bot API
AI Model	SEA-LION (via HTTP API)
LLM Prompting	LangChain + Custom Templates
Voice Input	speech_recognition (Sphinx) + pydub
TTS Response	gTTS (Google Text-to-Speech)
Mini App	Telegram WebApp + static HTML/JS
Hosting	Render.com (Webhook mode)
Storage	Redis or in-memory fallback

📁 Directory Structure

app/
│
├── telegram_bot.py       # Main app entrypoint
├── session_db.py         # Redis/memory store for user sessions
├── langchain_prompts.py  # Prompt formatting using LangChain
├── sea_lion_api.py       # Wrapper around SEA-LION HTTP API
├── templates/            # MiniApp HTML templates (optional)
├── static/               # JS/CSS assets for Mini App
└── .env                  # Secrets and environment variables

🚀 Deployment on Render

Set up a Python Web Service on Render.
Ensure webhook_url and TELEGRAM_TOKEN are in your .env.
Use the following startup command:

python set_webhook.py && python -m app.telegram_bot

Expose /telegram route for Telegram Webhook.
Add uptimerobot.com ping if you want to keep the bot alive.

🔊 Voice Input

Receives .ogg Telegram voice message
Converts to .wav using pydub
Transcribes with SpeechRecognition (Sphinx)
Uses text to query LLM, sends both text and voice reply (via gTTS)

🧠 Context & Memory

session_db.py handles storing user context (context:{chat_id}) and history (history:{chat_id})
Uses Redis if available, else falls back to in-memory Python dict
History truncated to last 4 exchanges (configurable via MAX_HISTORY_LEN)

🧩 Telegram Mini App

A simple WebApp interface in static/miniapp/index.html
Users can submit alerts (POST to Flask API)
Button appears via WebAppInfo inside /start reply

🧪 Testing & Debugging Tips

Test locally using python -m app.telegram_bot
Use logging.info(...) to track handler activity
Always wrap LLM calls in try/except for graceful failure
Redis not available? It will fallback to _cache in memory

🤝 Contributing

Extend with more wellbeing tools (e.g., journaling, CBT cards)
Swap SEA-LION with local Ollama models if needed
Improve multilingual support for wider accessibility

📄 License

MIT License – Use freely and improve for good 🌍

HappyBot is a capstone example of combining Generative AI, chat interfaces, multimodal input, and cloud deployment to solve real human needs.