Skip to main content

🔄 Data Flow

This page explains how data moves through the system — from a user's voice message to processed feedback and analytics.

1. Voice Message → FastAPI​

  • User sends a voice message to Telegram
  • Telegram forwards it to the FastAPI server (via webhook)
  • FastAPI pushes a processing task to Redis Stream

2. Audio Processing​

  • The FastStream worker listens to Redis and downloads the voice message
  • It sends the file to OpenAI Whisper for transcription
  • The transcript is enriched with feedback using GPT-4
  • Audio is stored in MinIO, transcript and feedback in PostgreSQL

3. Real-Time Analytics​

  • Debezium watches PostgreSQL changes and sends events to Redpanda
  • Bytewax ingests the stream and performs windowed analysis
  • (Planned) Results stored in ClickHouse and visualized via Grafana

To understand the architectural motivation behind this setup, see Challenges.