- 1
It arrives
A PDF, EPUB or scanned collection lands — uploaded by a librarian, or pulled in automatically from Kramerius and WebDAV with an LLM filling in the missing metadata.
- 2
Machines read it first
Before a human ever opens it, a fleet of independent workers goes through every page — pulling out text, OCR-ing scans, captioning figures, rebuilding equations as LaTeX, transcribing any video.
- 3
It becomes searchable
The extracted text is embedded into a vector index, so a Slovak query can surface a matching English or German passage — down to the sentence, next to classic keyword hits.
- 4
It can be lent, safely
If a title needs protecting, it’s encrypted with Readium LCP — the same open lending standard public libraries use — so it can be borrowed, renewed and returned.
- 5
The catalog holds it
Everything lands in the catalog at the centre — one install hosting many libraries — and is served over OPDS 1.2 and 2.0 through a single documented protocol.
- 6
Someone reads it
Students meet all of this as one thing: a portal they log into and a reader that opens any title in the browser — search inside, highlight, cite, borrow.
Why split it up like this?
The catalog stays small and dependable; everything expensive — OCR, captioning, embeddings, encryption — runs as its own worker on a job queue. A new capability is a new service, not a change to the thing holding the whole library. It also means each piece can be a student team’s project without risking the rest.
