Building My Personal Telegram AI Assistant: From Prototype to Production

The Challenge

Managing personal productivity across multiple tools is exhausting. Task managers, habit trackers, stock portfolio apps—I was drowning in context switching. I wanted a single interface where I could:

  • Create tasks naturally: "Remind me to call John tomorrow at 2pm"
  • Track health metrics: "I just did 30 minutes of cardio"
  • Monitor stock prices: "Alert me when IONQ hits $15"
  • Have intelligent conversations

So I built a personal Telegram bot powered by Claude AI.

Architecture Overview

Tech Stack

  • Framework: pyTelegramBotAPI (Python)
  • AI: Anthropic Claude API (natural language understanding)
  • Database: SQLite (self-hosted, 85% cost reduction vs PostgreSQL)
  • Deployment: Railway (persistent storage, auto-scaling)
  • Infrastructure as Code: Railway CLI

Current Features (v2.0.0)

📋 Task Management

  • Natural language parsing: "Remind me to submit expense report Friday at 9am"
  • Smart timeline recognition: "tomorrow", "in 30 minutes", "next Monday"
  • Persistent storage with auto-save
  • Interactive completion buttons

🏋️ Health Tracking

  • Protein logger with meal notes: /protein 150 eggs for breakfast
  • Workout tracker by type and duration: /workout cardio 45
  • Daily stats dashboard: Interactive metrics with trends
  • 7-day history with insights and patterns
  • Database statistics for tracking progress

📈 Stock Market Integration

  • Real-time price checking: /stock AAPL
  • Price alerts with directional triggers: /alert IONQ 15 below
  • Background monitoring every 5 minutes
  • Alert notifications when target prices hit
  • Persistent alert storage

🤖 AI-Powered Intelligence

  • Claude AI for intent recognition
  • Context-aware conversations with chat history
  • Multi-turn dialogue support
  • Fallback to general chat for non-task inquiries

Database Schema

Migrated from PostgreSQL to SQLite for cost efficiency:

users          → store user profiles and preferences
conversations  → chat history for context
health         → protein, workouts, daily metrics
preferences    → user settings and alert configurations
insights       → calculated trends and patterns

Development Journey

Phase 1: MVP (Task Management)

Got basic task creation and listing working with Claude AI parsing. Learned pyTelegramBotAPI routing and Telegram's button API.

Phase 2: Stock Alerts

Added real-time stock price monitoring, background tasks (APScheduler), and persistent JSON storage before moving to database.

Phase 3: Health Tracking & Database Infrastructure

Implemented SQLite schema, migrated from PostgreSQL, added health tracking endpoints. Cut operational costs dramatically.

Currently at ~800-1000 lines of monolithic Python. Time to refactor.

Next Steps (Roadmap)

v1.1.5 - Code Refactoring (In Progress)

Modularize into:

  • handlers/ - Command handlers by feature
  • services/ - API integrations and business logic
  • utils/ - Shared utilities (parsing, formatting)
  • models/ - Data structures

Benefits: easier debugging, parallel development, unit testing, cleaner git history.

v1.2 - Voice Assistant

  • OpenAI Whisper API for speech-to-text
  • Voice commands: "Remind me to call lawyer tomorrow at 2pm"
  • Support all existing command types via voice

v2.0+ - AI Enhancements

  • Stock sentiment analysis
  • Conversation summarization
  • Predictive reminders based on patterns
  • Multi-user support with permission levels

Key Lessons Learned

  1. SQLite > PostgreSQL for Personal Projects: No managed database costs, full control, easier backup
  2. Claude API is Powerful: Intent recognition with minimal prompt engineering
  3. Modular Architecture Matters Early: Starting with clear structure saves refactoring time later
  4. Railway is Perfect for This: GitHub auto-deployment, environment secrets, persistent volumes—set it and forget it
  5. Test in Telegram, Not Terminal: Real Telegram interactions reveal UX issues immediately

How to Get Started

# Clone and setup
git clone https://github.com/aeither/telegram-bot-python
cd telegram-bot-python
pip install uv
uv sync

# Configure .env
TELEGRAM_BOT_TOKEN=your_token
ANTHROPIC_API_KEY=your_api_key

# Run locally
uv run python -B main.py

# Deploy on Railway
railway init
railway up

Code Sample: Natural Language Task Parsing

# User sends: "Remind me to call John tomorrow at 2pm"
# Claude AI extracts intent, deadline, and creates task automatically

The bot is live and handling production traffic. Next month: voice commands. After that: open-source the modular version for other builders.

Want to build something similar? The modular refactor will be perfect for forking or extending. Follow the progress on GitHub.