How much model does the task need?
Retrieved-Span Training for Efficient Query-Focused Meeting Summarization on QMSum.
Read the paper
Build the future of homegrown AI.
We research and build custom AI models every team can use, train, and trust.
An applied AI research lab in Brunswick, Melbourne.
A model for every job.
Every business has work it knows deeply. We build specialist models around that knowledge, then refine them until they are as small and fast as the job allows.
Models trained on your task. Evaluated against the quality it needs. Shaped by your team.
How we work with enterprise teamsFrontier-level results, on your terms
Measured in our study on a summarization benchmark (QMSum).
Read the paper- 99%
lower cost than Claude Opus 5
and 96% lower than GPT-5.6 Luna
- +18%
better after training
ROUGE-1, same model before and after
- 23×
faster than Claude Opus 5
and 6× faster than GPT-5.6 Luna
A closer look. A wider possibility.
We start with a task your team knows well. Train on examples it can judge. Measure the result, then refine what the work needs.
Explore the endpoint programme
Our research
All researchModels that learn from traffic
How production requests and feedback become training data, and how a new version earns its place.
Agentic unit economics
Where AI spend goes, and what it costs to run agents at scale.
Read the analysisBuilt with care. Measured in the open.
Right-sized for the task.
Trained on your work.
Methods you can examine.