Bots
Content Intelligence Bot
A 24-hour digest of a creator's content, in Telegram
Pythonaiogram 3ARQPostgreSQLRedisWhisper STTOCRLLM function callingDocker
Problem
Understanding what a creator said in a day means watching every story and reel end to end. There's no text to read — the meaning is spread across speech, on-screen captions and overlays.
System
An async pipeline: content is fetched, speech is transcribed with Whisper, on-frame text is lifted by OCR, and an LLM with function calling folds it into a structured summary. Slow stages run in ARQ workers with state in PostgreSQL and the queue on Redis — the bot answers instantly while processing continues in the background.
Role
Solo build, commercial project.
Outcome
Dockerised, test-covered, ran in production for the client.