Models built, tuned and served by people who make their own.
Some jobs need more than a prompt on a general model. Leemer Group creates and adapts models for a defined task, prepares the data, hosts them where you choose and makes inference fast and affordable. The work draws on Leemer Labs, where the group trains, evaluates and releases its own models through the Born programme.
Model training & ML engineering
Custom and fine-tuned models, dataset work, hosting and fast inference, down to GPU kernels, from the team behind Leemer Labs and Born.
When is a custom or fine-tuned model worth it?
When a general model is too slow, too costly, too broad or cannot see your data. A smaller model trained for one task can be faster, cheaper and easier to run privately. We test whether that holds for your case before recommending it.
What does the ML engineering work cover?
Task definition, dataset collection and cleaning, synthetic and evaluation data, model creation, fine-tuning with full or adapter methods, distillation, quantisation, hosting and serving behind an OpenAI-compatible API.
Do you work on inference performance and kernels?
Yes. We profile serving end to end, then tune batching, caching and memory, and where it pays off design custom GPU kernels for the operations that dominate cost or latency.
What does Leemer Labs do?
Leemer Labs is the group’s AI research and model development activity. The Born programme brings model adaptation, evaluation with BornBench and held-out gates, and documented releases together, alongside fine-tuning and inference work.
How an engagement works.
- 01
Brief
A conversation about the problem, the people affected and what a good result looks like. You can draft the idea first in Build Brief.
- 02
Scope
A written scope: what we will build or test, what we will not, and how we will agree it is done.
- 03
Build
Focused work in short cycles. You see working software or evidence early, not a reveal at the end.
- 04
Handover
Code, documents and the reasoning behind decisions, in a form your team can run, change and own.
Questions, answered.
Can the model run on our own infrastructure?
- Yes. We can deploy on your hardware or cloud account, or through European infrastructure, and explain the trade-offs in cost and maintenance.
How do we know the model is better?
- Every model is measured against a held-out test set and against the general models it would replace, before release.
Do we own the model weights?
- Yes, for models trained for you, subject to the licences of any base model used. Licences are set out in the written scope.
How do we start?
- Use the contact page. It opens an email draft with this service selected, which you review and send yourself. No form collects your message on our side.
Related
services.
Leemer Group designs and builds AI assistants, agents and document search for a defined job, with private deployment when your data should stay with you.
AI evaluation & assuranceIndependent tests for an AI model or feature before you rely on it: task-specific test sets, failure analysis, vendor comparison and a written evidence report.
Irish-language & multilingual AIAI chat, translation and content for organisations that serve Irish speakers, evaluated with fluent speakers rather than assumed to work.
Tell us what
you’re working on.
Describe the problem and where you are now. We’ll reply with questions, not a sales script. Enquiries open as an email draft that you review and send yourself.