Six models. One job each.
Every tinyAI Labs model is small enough to run on a CPU, narrow enough to explain, and deterministic enough to sign off. Use them as they come, or train your own on about 200 examples.
Reads contracts, statements, registers and forms, works out what each one is, and pulls the fields you need into typed JSON. Every value comes with a confidence score and the line it came from.
IDs, scans, stamps, signatures, seals and inspection images. NETRA finds each region, labels it, checks scan quality and liveness, and tells you how sure it is.
Always on. Scores every transaction, covenant date and AML signal the same way at 3am as at noon, and attaches the rule and the reason to every flag.
Scores and decides with pinned, versioned weights. Replay any decision from any date and get the identical result, with the rule trace and attribution that produced it.
Matches what must match: owners against registers, names against sanctions lists, payments against invoices. Fuzzy where the data is messy, exact where the regulator is.
Train your own model on about 200 labelled examples, on a CPU, in under an hour. It ships with its model card and risk register, and you own the weights.
From your examples to production in an afternoon.
Bring ~200 examples
Your own documents and labels. No synthetic data, no GPU farm.
Train on a CPU
Under an hour from zero to a tested model, inside your environment.
Get the paperwork
Model card, risk register and conformity docs generated with the model.
Deploy anywhere
On-prem, private cloud or air-gapped. Built for EU AI Act, GDPR and DORA scrutiny.
Test any model on your own documents.
We run it inside your environment against your current approach. You keep the results and the model.