Initial import — HP Studios AI Content Agent
Full-stack app that turns HP customer briefs (master asset + regional
supporting docs) into a set of branded Word deliverables via a RAG +
agent pipeline.
Stack
- FastAPI + SQLAlchemy + pgvector + RQ (backend, Python 3.12)
- React + Vite + TypeScript + Tailwind + TanStack Query (frontend)
- Claude Opus 4.7 (generation) + Haiku 4.5 (translation/OCR)
- Voyage voyage-3 or OpenAI text-embedding-3-small (embeddings)
- python-docx (branded Word output, Montserrat + HP blue)
- Docker Compose (5 services)
Features
- 6 built-in deliverable types (leadership themes, regional enrichment,
LinkedIn posts, webinar spec, infographic specs, ABM enablement)
- Data-driven deliverable types: admins add new types at runtime via
prompt + JSON schema + template_json — no code, no deploy
- Generic schema-driven review form + generic Word template renderer
- Document ingestion pipeline with translation, chunking, pgvector RAG
- Pluggable auth provider (password now, Entra SSO later); admin/user roles
- Re-roll / retry on every deliverable; cascading delete; brief editing;
inline document upload; progress hints; router-level ErrorBoundary
- Admin panel with test-render preview for new deliverable types
- Help page at /help with architecture overview and usage guide
82 backend tests passing, 18 skipped (gated live-API tests).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
72c8a0d0fe