Forensic invoice auditing built on a hybrid intelligence engine — the speed of AI, the discipline of forensic accounting. The brand exists because no existing tool does both honestly.
LEGERIS exists to make forensic-grade invoice audit accessible, defensible, and repeatable. Accessible to organisations that could never afford a Big4 retainer for the work. Defensible because every finding is anchored to the exact line of the exact invoice with the math beside it. Repeatable because the same invoice through the same pipeline produces the same verdict, every time.
The hardest problem in AI-assisted audit is not the audit. It is the verification. Language models are good at proposing what might be wrong with an invoice. They are not good at being certain. They confabulate arithmetic. They cite contract clauses that don’t exist. They produce different findings on different runs against the same input.
Forensic auditors do not have that problem because forensic audit is not an interpretive practice. It is a deterministic one. You recompute the math. You verify the citation. You check the arithmetic identity. You ship the finding only when the structured evidence supports it. The audit holds up because the verification step is mechanical, not interpretive.
LEGERIS combines the two. Language models do the interpretation — reading contracts, proposing candidate findings, surfacing ambiguity. Purpose-built deterministic engines do the verification — recomputing every number, validating every citation, dropping any finding that doesn’t survive arithmetic scrutiny. The model proposes; the engines decide. The result is the throughput of AI with the evidentiary discipline of forensic accounting.
This is the architecture. Findings are the deliverable.
We do not let language models decide financial truth. Every numerical claim in a LEGERIS finding has been recomputed by purpose-built engines that are not capable of confabulation. If the engines and the model disagree, the engines win. Always.
A finding without its citation is an opinion. LEGERIS does not ship opinions. Every finding carries the source line — the page, the position, the value — so that the auditor reviewing it sees the proof, and the vendor receiving the dispute sees the proof.
Reproducibility is a discipline. The same invoice through the same pipeline produces the same findings, the same verdicts, the same report — no matter when it runs, no matter how many times. The audit hash printed on every report is the chain-of-custody mark.
Client invoice data, contract data, and audit findings are never used to train any model. Our subprocessor agreements contractually prohibit the language-model providers we use from training on data submitted through their API. This commitment is the floor, not the ceiling.
Founder & Lead Architect
Forensic accountants spend ninety percent of their time on the mechanics of audit — gathering evidence, recomputing numbers, formatting reports — and ten percent on the judgment that justifies their fee. AI inverts that ratio. LEGERIS exists to make forensic-grade invoice audit accessible to organisations that could never afford a Big4 retainer for the work, with the same evidentiary discipline a partner-led review would produce.
One engagement at a time. Scoped, quoted, contracted up front. No subscriptions. No surprises.