PIA Coach

[ Overview ]

PIA Coach is an AI accountability coach that runs on Telegram, built to fix one personal failure: owning a calendar full of meetings but zero discipline to execute on them. It connects to Google Calendar and Tasks, and builds a multidimensional behavioral model over time — tracking 21 event types, inferring 8 psychological states, and modeling circadian rhythms. PIA predicts when you are likely to procrastinate, adapts its coaching style (from drill sergeant to gentle nudge across 6 dimensions), and learns what actually works for you through LLM-scored intervention feedback. Named PIA — "Personal Intelligent Accompaniment."

[ Key challenges ]

• • Behavioral profiling from sparse event data — building reliable models from daily calendar interactions alone
• • Circadian rhythm inference without wearables — using calendar event timing patterns to estimate peak/slack hours
• • Psychological state inference (8 states: focused, distracted, anxious, burned out, etc.) from behavioral signal analysis
• • Adaptive coaching style across 6 dimensions (tone, frequency, directness, formality, empathy, accountability)
• • LLM-scored intervention effectiveness — closing the feedback loop to make the coach self-improving

[ Key features ]

• • Behavioral profiling across 21 behavioral event types — task completion, deferral, cancellation patterns
• • Circadian rhythm modeling to time coaching nudges at the most effective moments based on user energy patterns
• • Psychological state inference (8 states) from behavioral signal analysis with confidence scoring
• • Procrastination risk prediction using historical behavior, circadian phase, and current workload
• • Adaptive coaching style across 6 dimensions — adjust tone, frequency, and approach based on user responsiveness
• • Correlation mining — "detected: gym sessions improve next-day focus by 35%"
• • Tomorrow trajectory predictions — energy, execution, health, and momentum forecasts delivered each evening
• • Plugin-based architecture for extensible coaching modules and third-party integrations

[ Screenshots ]

[ Tech stack ]

• • Python backend with Telegram Bot API integration for conversational interface
• • DeepSeek LLM for personalized coaching conversations and psychological inference
• • Behavioral profiling algorithms — pattern recognition on event sequences for user modeling
• • Google Calendar API and Google Tasks API for calendar awareness and task synchronization
• • Circadian rhythm inference engine — ML model trained on temporal event distribution patterns
• • Plugin-based architecture for modular coaching features with typed interfaces
• • PostgreSQL for user profiles, behavioral logs, psychological state history, and coaching outcomes
• • Whisper (speech-to-text) for voice note transcription in Telegram messages