Creator-first Chapter Tool

A creator-first tool that turns raw speech into YouTube chapters. Transcription, an LLM pass for chapter extraction, and a human review step that puts the creator back in charge before publish. Open-source so creators can swap in their own ASR and model of choice.

I picked an open-source transcription pipeline over a hosted SaaS so the project stays portable: anyone can run it against Whisper, AWS Transcribe, or their own ASR with minimal glue. The language model needs clean text more than it needs a single vendor.

Chapter titles come from a structured prompt that prioritizes audience comprehension over keyword stuffing. The model proposes time-coded segments; the prompt enforces conciseness and consistent tense. Output is JSON the editor can edit before anything ships.

Chapters are never auto-published. The creator gets an editor that shows transcript, proposed timestamps, and proposed titles together, with one-click approve, inline edit, and drag-to-retime. The AI does the draft; the creator does the call.

Next.js for the editor surface and routing. A pluggable transcription service contract so any ASR backend works. An LLM API for chapter extraction with a typed JSON schema enforced at the prompt level. Session state in memory; nothing about a creator's drafts is persisted server-side without consent.