02/LuthorText/Doc AI, lending, insurance
Statements, invoices and claim forms, read by a model you host.
LuthorText pulls the fields out of the documents you see every day and sends only the odd ones to your LLM.
- Account•••• 4417
- PeriodMar 1 to Mar 31
- Deposits$84,210.55
- Closing balance$31,902.18
In plain words
If your product reads documents, you probably pay an LLM for every page. Most pages follow a layout you have already seen.
LuthorText is trained on your document types and reads those pages by itself, on your own servers. It works on the same idea as LuthorSQL, our SQL model: answer what it is sure about and pass on the rest. A skewed scan or a handwritten note still goes to your LLM.
How it works
- 01
A document arrives.
A bank statement, an invoice, a claim form, a pay stub.
- 02
LuthorText reads it and scores itself.
It returns the fields and how sure it is about each one.
- 03
Anything it is unsure of goes to your LLM.
Hard pages go to the LLM you already use.
What you get
- LuthorText, trained on your document types and your field names
- The same output format your LLM returns now
- Checks that totals add up, so obvious tampering gets flagged
- Runs in your cloud account or on your own hardware
The honest part
We have not published numbers for LuthorText yet. The measured results on this site come from LuthorSQL, our SQL model. The savings on your documents are an estimate until we measure them, which is what the first week is for.
See what LuthorSQL or LuthorText could take off your bill.
Fifteen minutes. Bring a rough idea of what you spend on LLM calls each month.