Fabel5
Task authoring and evaluation against a working engineer's judgement rather than a checklist.
Services · Last six months
We work with model teams.
Contractual engineering for the organisations leading the field, including Anthropic, OpenAI and Gemini. Six continuous months across three models, trained with advanced reinforcement learning techniques and reviewed against each team’s own quality bar.
3Model programmes
6 moContinuous engagement
Models we have trained
Three programmes, six months
Mythos
Engineering work delivered to spec, reviewed and iterated against real model behaviour.
Gemini
Continuous contribution across the programme, scaled up and down as the work required.
How an engagement runs
Four steps, no lock-in
Scope
You describe the work. We say plainly what we can take on and what we cannot.
Trial slice
A small piece delivered first, so both sides see the standard before committing.
Deliver
Continuous contribution at an agreed rate, reviewed against your own quality bar.
Scale
Widen the engagement or wind it down. No lock-in either way.
The work itself
Fabel5MythosGeminiWhat we do
Engineering delivered under contract to model teams: the tasks that need a working engineer's judgement rather than a checklist, and the evaluation work that shows whether an output is actually better.
Who we have worked with
Over the last six months our work has covered the Fabel5, Mythos and Gemini programmes, across authoring, review and evaluation.
How to engage us
Short scoped engagements or a continuing arrangement. Tell us the shape of the work and we will say what we can take on and when.
Have work that needs an engineer?
Tell us the shape of it. We will say what we can take on, at what rate, and when we could start.