Logit Logic is a small applied-research practice. We design and build machine learning systems to order, for organisations whose problems are not served by a product.
The discipline does not change between engagements: a precise statement of the problem, models whose behaviour can be measured and bounded, and engineering to the standard of the system they will run in. The domain is whatever the problem requires.

Practice
- Engagements are bespoke and scoped to a single problem. There is no platform and no product.
- Work runs from formulation to deployment: data, modelling, evaluation, and integration with the systems the result must live in.
- Known structure is used where it exists, whether physical law or domain constraint. Uncertainty is estimated and calibrated, not assumed.
- Evaluation is designed before the model, including how it fails.
- The practice is deliberately small. Those who scope the work carry it out.
Domains
One discipline, applied where the problem is. The list is representative, not exhaustive.
| Aerospace and propulsion | Dynamic models of turbine performance and fuel efficiency, fast enough to run inside the loop. |
|---|---|
| Energy | Optimisation of power delivery across constrained networks. |
| Healthcare | Systems that operate on protected health information. |
| Security | Offensive security assessment and penetration testing. Machine learning built to withstand adversarial use. |
| Agent systems | Custom agent harnesses and autonomous tooling, with bounded authority and a complete record of action. |
| Government and defence | Work with sensitive and controlled data. |
| Intellectual property | Models built on proprietary and IP-sensitive material without exposing it. |
Sensitive data
Much of our work involves data that cannot leave its owner's control: protected health information, proprietary and unpublished intellectual property, and government or tactical data.
- Engagements are conducted under non-disclosure agreements. We do not name clients or describe their work.
- Work is carried out in isolated environments: on the client's own infrastructure where possible, otherwise on dedicated, access-controlled systems provisioned for a single engagement.
- Client data, and models derived from it, are used only for the engagement in which they were provided, and are returned or destroyed at its close.
- Handling requirements specific to an engagement are agreed in writing before work begins.
Contact
We take on a small number of engagements at a time. Enquiries by email:
A brief description of the problem is sufficient. Please do not include sensitive material in a first message; we will arrange a suitable channel.