Most software for operational industries is built by people who have never done the operation. It shows. The workflows are close but not right, the vocabulary is almost correct, and the parts that matter most — the edge cases that actually consume a working day — are the parts nobody modelled.
Devinity Solutions was set up to build the other kind.
What Devinity builds
Devinity is a software company working on intelligent systems for businesses whose value is created away from a keyboard — logistics, field operations, service delivery. The common thread is that the important information arrives as documents, calls and physical events rather than as tidy form submissions, and somebody is doing the translation by hand.
That work divides into three things:
- Applied AI systems — document understanding, extraction and classification pipelines that turn unstructured paperwork into structured records a business can act on.
- AI agents — software that carries out multi-step operational work end to end, with a human reviewing the output rather than assembling it.
- Operational platforms — the multi-tenant, role-scoped systems underneath: permissions, audit trails, financial calculation, reporting and the unglamorous correctness that decides whether people trust the numbers.
The last one is the part that gets skipped most often and matters most. An extraction model that is right nine times in ten is a demo. The same model inside a system that flags its own uncertainty, routes it for review, records who changed what, and keeps the arithmetic consistent across a whole account is a product.
The method: build it for your own desk first
Devinity Dispatch — the company's own dispatch operation — books freight for American carriers. It handles rate confirmations, negotiates rates, files paperwork, invoices brokers and chases payment. It is a real operation with real deadlines and real consequences for getting a number wrong.
fleetchart was built to run that desk. Not as a pilot or a proof of concept — as the system the business depended on. Every feature had to survive contact with a Friday afternoon and a broker disputing detention.
That constraint produces different software than a roadmap does. Features that demo well but slow the desk down get removed. The unglamorous things — a document that will not parse, a correction that has to stick, a permission boundary between a driver and a financial field — get built properly, because somebody in the building hits them every day.
What that looks like in fleetchart
The visible layer is document understanding: a rate confirmation PDF goes in, and rate, miles, broker, origin, destination, dates and load number come out as structured data, with anything unreadable flagged rather than guessed. Extending that reading to receipts and bills of lading is the next piece of work rather than a shipped feature.
The layer underneath is what makes it usable. Extraction is presented for review rather than committed silently. Uncertain fields are flagged instead of guessed. Corrections stand. Access is scoped by role, so a driver sees their own loads without financial fields and a dispatcher sees only the trucks assigned to them. Each carrier's records are isolated from every other carrier's at the application layer.
None of that is exciting to describe. All of it is the difference between a financial record you can act on and a dashboard you quietly stop trusting.
Why AI agents matter here
Extraction removes typing. Agents remove the loop around the typing — the chasing, the checking, the routine follow-up that consumes a back office.
In a trucking context that is a large surface: confirming a document package is complete before an invoice goes out, noticing that a broker's payment is overdue against their own historical pattern, flagging a lane whose margin has drifted, reconciling a fuel card statement against the receipts already on file. Each is small, repetitive, rule-bound work that a person currently does badly because they are also driving.
The engineering discipline is knowing where an agent should act and where it should only recommend. Money and compliance are places to recommend. Devinity builds to that line deliberately, because the alternative — software that quietly does the wrong thing at scale — is worse than software that asks.
Working with Devinity
fleetchart is the product you can see. The underlying capability — document understanding, agent workflows and the operational platforms that hold them — is what Devinity builds for operational businesses more generally.
If you run an operation where the real information arrives as paperwork and somebody is retyping it, that is the problem this company exists to remove.