Evidence-linked heritage

Old manuscripts.
New scrutiny.

We scan primary sources, structure every claim they contain into a queryable evidence graph, and check it against the scholarship — so history stops resting on somebody's summary and starts resting on the page it came from.

320
Entities extracted
285
Cited relationships
150+
Years of debate checked
Why this exists

Don't summarize
the archive.
Interrogate it.

Most AI applied to history does one thing: it reads an old text and tells you what it says. That's a summary — convenient, but no more trustworthy than whoever wrote it.

Pamet takes the harder path. We scan the original pages, extract every entity, claim, and relationship they contain with a page citation attached, and build the result into a graph you can walk through yourself: click a name, see every place it appears, every piece of evidence for and against what's claimed about it, and where scholars still disagree.

Nothing is asserted without a citation. Nothing is settled without showing its work.

01
Digitize
Full-resolution scans, OCR'd page by page from the primary source.
02
Extract
Entities, relationships, claims — each tied to an exact page and quote.
03
Normalize
Free-text extraction bucketed into a browsable, controlled vocabulary.
04
Cross-check
Checked against published scholarship; every claim gets a confidence level.

Read the full method →

More cases, one method

Case studies

Veda Slovena is the first entry. The method doesn't care what shelf the manuscript came from.

"Don't build AI that summarizes books.
Build AI that explores evidence."

Pamet — project thesis