Video Creator
An automated pipeline that turns an article into narration, assembled video, and multi-platform uploads.
Read the build from several angles.
Authentic screens, terminal output, and architecture plates are kept together so the page shows the shape of the work without inflating the claim.
This record uses a procedural architecture view because no authentic public screenshot has been approved.
AI
What exists.
Video Creator automates the journey from Reddit stories or articles to narrated, assembled videos with direct uploads to YouTube, TikTok, and Facebook.
Where it resisted.
Video generation involves research, scripting, audio synthesis, visual assembly, and platform-specific upload formats. Each step has its own failure modes.
What I changed.
Built a monorepo with a CLI core, SvelteKit web interface, and Express OAuth proxy for secure platform authentication.
The obvious implementation was not the product.
Use a CLI-first approach with a web interface on top, so the pipeline can run headless in automation.
Current proof
03- CLI and web interface for story browsing and video generation.
- Multi-platform uploader with OAuth proxy.
- Remotion rendering pipeline with AI scene boards.
Public detail is intentionally limited. No private source, customer data, or sensitive architecture is exposed.
What this build changed.
Automated content pipelines are fragile at every boundary: the value is in making each failure recoverable.
The implementation surface.
Proof should be inspectable.
Links are intentionally withheld for this private working set.