MACHINE LEGIBILITY
CONTACT ↴
001/THESIS

Most of the world's knowledge is illegible to machines.

Reports, transcripts, spreadsheets, folder names, project files, someone's memory of a call. It is all sitting in formats software cannot read. AI is only as good as what it can see, and right now it cannot see most of this. We make it legible, then build on it.

↓SCROLL TO RESOLVE
002/WHAT WE DO

We make real work legible to machines, then build on it.

01
Translate the work into data.
ReadDocuments, records, project files, whatever a domain actually produces. We turn it into something software can read, where it already sits.
02
Build the software on top.
BuildCustom tools and AI agents built around one domain and how it really works. Some of it does the work. Some of it just understands the work well enough to be worth asking.
03
Keep people where people matter.
IntegrateWe find the judgment that should stay with a person, and build everything else around it.

Most AI right now is pointed at work it cannot see. We fix the input first. We do that two ways: on engagement, inside a client's operation, and on our own, as products we build and ship. Same practice either way.

003/THE CONDITION

Every tool you bought added one more place for the truth to hide.

A
A forty-page report nobody has read since it was submitted.
ContainsThe only written record of what the project actually is, buried on page 27, in prose.
B
Eleven versions of the same spreadsheet.
ContainsThe current numbers, in exactly one of them. The filename does not tell you which, and the person who would know is on leave.
C
A meeting that happened.
ContainsThe decision. Agreed out loud by everyone present, written down nowhere, quietly contradicted six weeks later.

Filing it better does not help. Something has to be able to read it.

004/DEMONSTRATION

One document, two states. Drag across it.

DOC-4417 · UNSTRUCTURED SOURCE PARSED 000%
record.typemilestone_update
project.idPRJ-0417
project.statusAT_RISK
milestone.q2slipped · +90d
milestone.evidenceemail · unratified
budget.currentrev_3 · 2026-05-14
budget.supersededdrive/shared/v2.xlsx
personnel.piunchanged
vendor.engagementterminated
decision.ownerunassigned
contradictsDOC-4102 · DOC-3988
extraction.confidence0.91
source.locationunchanged · in place
← DRAG · SAME DOCUMENT, TWO STATES → NOTHING WAS MOVED, RENAMED, OR MIGRATED

Left is what your organization looks like to software today. Right is what it looks like after we're done. The document never moved. Same folder, same name, same owner. It can just be asked a question now.

005/CONTACT

If your institution's memory is trapped in documents, write to us.

will@getmachinelegibility.com
Machine Legibility LLCChapel Hill, NC 27516
United States
>
Monday to Friday, 9:00 to 17:00 ET
Good fit for engagementsOrganizations that run on documents and institutional memory. Research commercialization, grantmaking, economic development, anyone administering a portfolio of things they didn't build.
Bad fitAnyone shopping for a proof of concept, a pilot to satisfy a mandate, or a vendor to sit at the far end of a procurement process.
How it startsOne conversation about the specific thing you keep having to explain to people. If we can't see the system in it within an hour, we'll tell you.