Knowledge base to published content, with review lenses and a learning loop.
A rights-aware platform that turns openly licensed books into learning articles — one site per book, under a parent hub.
Most tools that turn a document into articles begin with the writing. This one begins with the right to publish at all. A source is classified before anything is drafted, and the licence rules are enforced as database constraints rather than application logic — a non-commercial licence cannot be relabelled as permissive, because a foreign key forbids it.
Extraction keeps code and tables byte-for-byte and unmerged, so their fidelity can later be scored by exact comparison instead of asking a model to grade itself. Every fragment carries the page range it truly came from.
This service is the front door for the phases marked built. Nothing below them is running, and this page will say so until it is.
Schema, rights engine, guarded model client
21 tables, 10 enums, 15/15 database gates. India life+60.
Extraction and the unit planner
PDF, EPUB and HTML to fragments carrying true page ranges; a 300-page book plans into ~25 units.
Measurement harness and the five scores
Extraction accuracy, code and table fidelity, hallucination rate, editorial effort, originality.
Generation and deterministic verification
Drafting, claim-support checking, verbatim-overlap gate.
API, durable worker, review dashboard
This service is its front door.
Static renderer and the published sites
Deployment, TLS, DNS
Firewall, backups with a tested restore, monitoring
Marks, wear and shelf; the learning loop
POST /api/rights/classifyClassify a licence and say whether it permits generation.
POST /api/extractExtract a PDF, EPUB or HTML document into fragments and plan it into units.
GET /api/statusLive build state and database facts.