
A fully engineered AI-powered Telegram coaching system designed to help users improve decision-making, accountability, and emotional regulation through structured conversations and performance tracking.
The platform combines conversational AI with a robust backend architecture, enabling real-time coaching, weekly accountability audits, and intelligent progress tracking. Users interact naturally (no commands), while the system handles subscription gating, crisis detection, and personalized coaching flows.
Key features include a state-driven coaching engine (Diagnose → Reframe → Activate → Regulate), weekly commitment tracking, and a “Read-to-Coach” progress summary that transforms user data into structured, spoken accountability reports.
The system integrates OpenAI (Assistants API + web search fallback), Supabase (PostgreSQL), and Stripe for subscriptions, ensuring scalability and production reliability. It also includes advanced handling for edge cases such as crisis scenarios, rate limiting, and failure recovery.
Additionally, voice support was implemented, allowing users to send voice messages, which are transcribed, processed through the AI pipeline, and returned as voice responses creating a fully conversational experience.
This project demonstrates strong capabilities in backend architecture, AI integration, system design, and real-world production deployment.