OSKETH
OSKETH ⟡ DOCUMENT INTELLIGENCE SYSTEMS

Built to know
when it isn’t sure.

Osketh builds quiet software for regulated back offices — the desks where somebody opens a PDF on one screen, types what it says into another, and checks it twice, because a wrong number means a denied claim, a failed filing, or a check that doesn’t go out. We read the document, check every value against the rules that govern it, and file the result where it belongs.

⟡ 01 — What we build

Small, sharp tools for heavy paperwork.

SYS.01

Read

Scans, faxes, portal downloads, forty-page PDFs. Our pipelines turn messy documents into structured records — every value traced back to the exact spot on the page it came from.

SYS.02

Check

Codes must exist. Columns must sum. Rates must match the published tables. We verify with hard rules first, models second — and route the rare exception to a person, not the other way around.

SYS.03

File

Approved records post into your system of record — preview before, undo after, an audit trail throughout. The work ends where your books actually live.

⟡ 02 — Organization

Built for the places where paperwork moves money.

ORG.01

The Company

Osketh builds narrow, deterministic software for regulated US back offices. We take high-consequence document workflows, automate the reading and verification, and file the finished records into systems of record.

ORG.02

The Philosophy

Language models do the reading; published rules and official tables do the judging. The system is designed to know when it isn’t sure, making the human review cheap, exact, and provable.

ORG.03

The Scope

Focused to begin with on three regulated document families: claims and remittances, certified payroll, and compliance filings. Small, sharp tools with an audit trail throughout.

  accuracy you can audit   humans where judgment lives   built for regulated work   quiet software, loud paperwork
⟡ 03 — Status

NOW ⟶ building the first products with early partners, in the places where documents move money: claims and remittances, certified payroll, compliance filings. US verticals. Deliberately unglamorous.
NEXT ⟶ announced here, quietly, when they're ready.

⟡ 04 — Reference

Notes on the paperwork.

Certified payroll electronic filing, state by state

What California, New York and Washington each require — formats, schemas, filing cadence and penalties — alongside the federal weekly baseline, and why one file can't serve all three.

EOB vs ERA (835): what's actually in each

A plain-English glossary of remittance terms: who each document is for, the segments inside an 835, what CARCs and RARCs do, and the arithmetic that has to balance.

WH-347, field by field

Every field on the January 2025 revision of the federal certified payroll form, from the Wage and Hour Division's own instructions — and the rules inside them that are easiest to miss.

⟡ 05 — Questions

The obvious questions.

What is Osketh?

Osketh is a document-intelligence company building single-purpose software for regulated back offices in the United States — systems that read a document, check every value against the rules that govern it, and file the result where your books actually live.

Which documents does Osketh work on?

High-consequence paperwork in regulated US workflows, and each family has its own way of being wrong.

Claims and remittances — every adjustment code must exist, the line amounts must add up to what was paid, and the balance left over must be the one the payer says it is.

Certified payroll — every classification must carry a rate at or above the published wage determination, fringe must be accounted for, and the hours must reconcile week to week.

Compliance filings — every field the form requires must be present, the totals must match the period they cover, and the form must be the version currently in force.

How does Osketh know the data it extracted is correct?

Rules first, models second. Language models do the reading; deterministic rules do the judging. Every extracted value keeps a pointer to the exact spot on the page it came from, so a reviewer checks it against the page instead of hunting for it. A value with nothing behind it is suspect by construction.

What happens when a document is ambiguous?

It goes to a person. Any value that fails a check is routed to a review queue with the source document pinned to the flagged field. The system is built to know when it isn’t sure, rather than to pretend it always is.

Where does the finished record end up?

In your system of record. Approved records post with a preview before, an undo after, and an audit trail throughout. Nothing is written that you haven’t seen first.

What happens to our documents?

They stay yours. We do not train models on customer documents, and we do not sell or share them. Retention is written into the agreement — a fixed window, and deletion when you ask. What we hold, how long we hold it, and who can open it is settled in writing before the first document moves.

Can I use Osketh today?

Not yet. We’re building the first products with early partners, and nothing is launched publicly. New products are announced here when they’re ready — if your team retypes documents that must be perfect, get in touch.

⟡ 06 — Contact

Talk to us.

If your team retypes documents that must be perfect — or you're building in the same direction and want to compare notes — we read everything that comes through this form.

prefer email? [email protected] · what we do with what you send: privacy