Content pipeline service
A scheduled backend pipeline that plans, renders and publishes short videos end to end, built as small restartable stages with idempotent uploads. In progress.
- 5restartable stages
- 0duplicate uploads by design
What I’m building
A backend job that turns a topic into a finished, uploaded video every day without anyone touching it: plan the content, generate assets, add narration, render with FFmpeg and publish through the YouTube Data API. The first version runs on one machine on a schedule; the current work is making it portable.
How it is put together
- Stages with files in between. Each stage writes its output (plan, images, audio, video, state) to a dated folder, so any stage can be rerun alone instead of redoing the whole job.
- Idempotent publishing. The job records the video ID after upload and refuses to retry blindly if a previous run stopped mid-upload, so a network failure can never post twice.
- Scheduled like a service. A macOS LaunchAgent runs it daily with logs to disk; OAuth tokens and settings live outside the repository.
- Local generation, no API bill. Content and images come from local models today, so a daily run costs nothing beyond the machine it runs on.
What’s next
Add the hosted-model adapters, package the pipeline in a container, move the schedule to a cloud runner, and add health checks and alerts so a failed day is noticed. This is where I’m practising the DevOps side: deployment, scheduling, secrets and monitoring.