Postgres-first, governed, and degradable
Maestro is engineered to run in real institutional environments — a single required source of truth, optional accelerators, and a pipeline that keeps humans in control.
Data architecture
One source of truth, clear boundaries
Postgres + pgvector
The required primary store for entities, auth, audit, pipeline artifacts, references, and embeddings.
Graph projection
An optional secondary graph for traversal and visualization. The core runs without it.
Filesystem assets
Uploaded syllabi, references, avatars, and rendered video live on disk with metadata in Postgres.
Retrieval
pgvector powers grounded reference retrieval, coverage, and alignment checks.
Project shape
A clean split between engine and experience
The backend owns the governed pipeline and data contracts; the frontend renders Studio, Journey, and admin surfaces.
Project structure
backend/
auth/ JWT auth, RBAC, permissions, course access
db/ Schema, repositories, migrations (Postgres + pgvector)
node-engine/ Node sets, blueprints, content specs, modalities
routes/ Express route families
services/ Course stages, council, references, library, audit
models/ Shared data contracts
frontend/
components/ Studio, Journey, library, and UI components
pages/ Dashboard, course, library, admin
services/ API client and auth helpersAI execution
Single model or a council
Run one model per stage, or let multiple members respond and a chairman model synthesize. Node Engine prompt edits create immutable new versions; produced objects keep the exact version used.
Governance
Auditable by construction
JWT auth and role-based access (admin, professor, student), course ownership, reviewer assignments, and an audit log gate the pipeline end to end.