Jun 2025
Multi-Model Video Processor
Video pipeline with transcription and frame extraction
FastAPIWhisperPostgreSQL
Problem
Getting structured data (transcripts, key frames) out of arbitrary video — whether uploaded directly or pulled from YouTube — usually means stitching together several different tools by hand each time.
Architecture
FastAPI orchestrates the pipeline: ingest (upload or YouTube), Whisper for transcription, a frame-extraction step, and Postgres to persist the structured output per video.
Key decisions
- —Whisper for transcription rather than a cloud speech API — runs locally without a per-call cost, which matters for processing more than a handful of videos.
Still being written up
Repo cleanup still pending: several self-generated status-report markdown files and ad-hoc test scripts are sitting at the repo root instead of a tests/ folder. No live demo currently.