project creatortrack scripts voice profile project 20260715T173150
name: CreatorTrack Scripts app: voice-profile script generation build (phase 1+2) description: CreatorTrack building an AI script-generation pipeline (515 YouTube transcripts -> structure extraction -> retention-weighted voice profile -> generation) plus IG/FB/TikTok ingest with cross-post matching; ships in two phases with a documented productization ledger. type: project
Two parallel branches building CreatorTrack's Scripts app for Gus Outdoor content:
- Main lane (feat/social-data-layer-and-scripts): YouTube transcript ingest -> per-video structure extraction (hook/beats/CTA/tone via Haiku) -> retention-weighted voice profile synthesis (Sonnet) -> pgvector embedding + few-shot script generation -> Scripts app UI + CT chat tool -> blind holdout eval as the quality gate.
- Meta lane (feat/social-meta-ingest): IG/FB/TikTok video ingest + cross-post matcher (links reposts across platforms to the source YouTube video) + UI chips.
- Ship order: phase 1 (main lane) ships first via /ship, phase 2 rebases onto it and ships second. Requires Justin to reconnect YouTube OAuth on workspace 85 post-ship for impressions/CTR.
Productization ledger (data architecture is multi-tenant/workspace-scoped by design already; four Justin-only shortcuts remain before other users can use this):
1. Meta system-user token (Justin's Business Manager pre-auth) -> needs Facebook Login for Business OAuth + Meta App Review per scope.
2. TikTok sandbox app (Gus is the registered test user) -> needs TikTok production audit (5-10 days, demo video).
3. Transcripts via yt-dlp on one residential IP -> doesn't scale per-user; likely answer is per-user OAuth + captions API, or Whisper for caption-less creators. Flagged as the biggest open product question.
4. LLM calls billed to Justin's Claude subscription -> needs metered billing; CT already has BYO-key infra (user_llm_keys, encrypted) that the extraction/profile/generation jobs need to wire into per-workspace.
Why: Justin asked directly whether this build would be duplicable for other users one day; this ledger is the answer and the future phase-3 scope. How to apply: when picking up phase-3 work on CreatorTrack's social/voice pipeline, or when a new user asks to use Scripts app, start from this ledger rather than re-deriving the gaps.
[auto-memory session 60d132de-4659-485c-8044-427af0ccce7f, confidence 0.72, mode staged]