SCANOVICHApplied AI research lab
Projects/prototype

Document-intelligence core

One engine loop for every document domain — a new domain adds its own validators and precedents, not a new system.

Status
prototype
Code
closed code
Research line
Documents that carry liability
Updated
2026-11-11

Ingest. Extract and normalise. Retrieve precedent. Produce an explainable structured output. A person signs. Deploy privately if needed.

What works today

  • A library that implements the loop without any domain built in.
  • A domain plugs in as a description of its corpus, its routing patterns and a golden set of examples.
  • It runs on the lab’s local embedding and reranking services.

Why it is in the lab

Trade documents are the first domain. The question is how cheaply the same core transfers to a second and a third.

How to engage

A demonstration on your own documents — messy inputs in, a verifiable record out.

Discuss this project