Leveret SystemsYour evidence already exists — finding it again should not be a project
In regulated manufacturing the cost is rarely the testing. It is the search — every change, every variant, every new quote sends engineers hunting through prior submissions, standards and test reports for something they proved a year ago. We build a cited rules engine around the systems you already run: evidence drawn through your systems' APIs into a governed warehouse, processed there, and returned with the governing clause attached — to your existing tools where integration allows, or to an interface built to fit the process where it does not. No rip-and-replace, and your systems stay the source of record.
For regulated manufacturers
Stop re-proving certification evidence you already hold
Engineers lose hours hunting through prior submissions, standards and test reports for evidence that already exists. We build an assistant that answers from your own standards and your own past certification evidence — every answer cited to the governing clause, with a deterministic rules-engine gate in front of the generative layer. Deterministic where compliance demands it, AI-assisted where it saves time, human-signed-off throughout. A day or two on site for discovery; the build itself runs remotely in your own tenant. And the destination is bigger: bind the rules to the parametric values in your CAD models, and a configuration is checked as it is designed — the doorway to generative design within compliant bounds.
It works from where your evidence already lives: drawn through your systems' APIs into a governed warehouse, processed there, and returned to your existing systems — or surfaced in an interface built to fit the process where direct integration is not possible. It starts with a four-day audit on one product family: at the end you have a findings pack and you have seen the engine running on your own data, so the decision about going further is made on evidence rather than on a proposal. The integration build is the usual next step, priced separately — with the audit fee credited against it if you go ahead within ninety days.
Your data stays yours. Engagements run inside your own cloud tenant wherever possible; otherwise in a UK-resident, single-client environment stood up for you alone. Your documents are never used to train models. At the end of the engagement everything is handed over or deleted — your choice, in writing.
What we build
What it takes to reach your evidence
A rules engine is only as good as what it can read. Certification evidence sits in a PLM, a document store, an ERP and a set of CAD models that were never designed to talk to each other — so the integration is the hard part, and it is the part we do. Four layers, one build.
Rules engine & cited retrieval
A deterministic rules engine computes what a change triggers and what evidence closes it, clause by clause. Retrieval answers from your own standards and prior certification records, with every claim traced to its source passage. Where a fact is missing the answer is undetermined and a named person is asked — nothing is guessed, and nothing is approved by a machine.
Integrated with the systems you already run
APIs first: evidence is drawn from your PLM, ERP and document store into a governed warehouse — typically BigQuery — and processed there. Results go back into your systems where integration allows; where a third-party system has no API, we build an interface that sits seamlessly in the existing process instead. Your systems remain the source of record.
CAD & engineering data
The model's parametric values are read directly, so a configuration is checked as it is designed rather than caught at review. Autodesk APS and SolidWorks pipelines, bill-of-materials structure, and engineering-change traceability that keeps as-built matching as-designed.
Data foundations & reporting
Where the thread does not exist yet, we build it: the APIs and data flows that join your systems into one model, with reporting on BigQuery and Power BI. This is the plumbing that lets the engine read everything above — built once, owned by you. And where existing capture cannot yet support the engine, we say so at the audit and scope that development on its own.
Why Leveret
Built in, not bought
No rip-and-replace
Most compliance tools ask you to re-key your evidence into them, which is why so many are quietly abandoned within a year. We work the other way: evidence is collected through your systems' APIs into a governed warehouse, processed there, and returned to your existing systems — or to an interface built to fit the process where direct integration is not possible. Your systems stay the source of record.
Deterministic, cited, signed by a human
Generative components sit between deterministic gates. What a configuration triggers is computed, not guessed; every answer carries its governing clause; and where a fact is missing the engine returns undetermined and escalates to a named person. It never approves anything.
Maintained as regulation moves
Standard editions are superseded, designated lists change, exemptions are granted and refused. Rules that are not maintained stop being true — so maintenance is a standing part of the service, priced per rules engine, rather than an afterthought.
About
Leveret Systems
Leveret Systems has been operating since 2014. We specialise in deterministic systems that use AI for cutting-edge classification: rules engines that compute what a change triggers, retrieval that answers with the governing clause attached, and classifiers that trace every result back to its source. Deterministic where compliance demands it, generative only where it genuinely saves time.
We build these systems around regulated manufacturers' existing infrastructure: integrating directly where a system offers an API, and building interfaces that fit the existing process where one does not. Engagements start with a fixed-price, four-day audit on one product family, grow through integration, and continue as rules maintenance, because regulation moves and rules that are not maintained stop being true. Where a build needs more hands, we scale to suit; the model does not change — one accountable partner, everything cited.
Selected work
Case Studies
The digital thread in production. Open any one for the full breakdown — the first is the clause-cited compliance assistant that the certification-evidence engine is built on; the others connect CAD, ERP-class data and BI on one model.
AI engineering
How we keep AI consistent in production
Architectural principles
Engagement
Pricing
Most engagements start with a fixed-price, four-day audit at £3,500 on one product family — credited in full against the integration build if you commission it within ninety days. At the end of the four days you have a findings pack and you have seen the engine running on your own data, so the decision about going further is made on evidence rather than on a proposal. From there the work is integration: direct through APIs where your systems allow it, through purpose-built interfaces where they do not. The audit also gives you a straight answer on whether your data can support a rules engine at all — where existing capture or storage falls short, the development that closes that gap is scoped and priced as its own workstream, because it is closer to R&D than integration. And because regulation moves (standard editions are superseded, designated lists change, exemptions are granted and refused) every installed rules engine needs maintaining, which is what the monthly tier is for: priced per engine, so a second rulebook carries its own. Capacity flexes; the rate does not.
Engagement shapes
Day rate
One rate for every kind of work — architecture, CAD and ERP integration, rules-engine and applied-AI build, data and BI. There is no junior tier and no blended rate, because there is nobody to blend with: the person who scopes the work is the person who does it. The rate does not escalate between stages, and it does not change with the discipline you need that week.
Who it's for
Where the digital thread pays off
Aerospace & Cabin Interiors
Certified programmes: CAD-driven configuration, airworthiness/TSO evidence, weight roll-up and as-built traceability.
Precision & Composite Manufacturing
Engineering BOM to manufacturing BOM with no re-keying; engineering-change control and shop-floor data capture.
Industrial Equipment & Machinery
Design-to-build-to-service on one data model; spares, warranty and in-service data turned into a profit centre.
Medical Devices
Technical files and design history: cited retrieval across standards, test evidence and prior submissions; traceability from requirement to as-built.
ATEX & Pressure Equipment
Conformity evidence: find and re-use prior assessments, cited to clause; keep the technical file consistent with the CAD model.
Rail & Marine
Certification and survey evidence: one searchable, cited view of standards, approvals and past submissions for each product family.
Get in touch
Start a conversation
Tell us which product family eats the most certification-evidence hours, and you will get a straight answer on whether the four-day audit fits. Or, for the wider digital thread, just start the conversation.