05 · Model integration & fine-tuning
The right model — chosen on evidence.
New models arrive every month. We pick on measured results for your task — quality, cost, latency and where the data goes — integrate behind a clean interface, and fine-tune only when the numbers say it’s worth it.
Start with your use case
Choose what matters most and we’ll show the kind of model setup we’d usually evaluate first.
Options we work with
Hosted, private or both.
We’re not tied to a vendor. The right answer is often a mix: a capable hosted model for hard requests and a smaller private model for sensitive or high-volume work.
Leading hosted models
Managed APIs from the major providers, configured so your data isn’t used for training, with Australian-region options where offered.
Open-weight, self-hosted
Open-weight models running in your cloud tenancy or infrastructure, when data must not leave an environment you control.
Fine-tuned models
Smaller models trained on your examples for specialised formats, terminology or tone — faster and cheaper per request once trained.
When to fine-tune
Fine-tuning is the last step, not the first.
Most quality problems are solved by better retrieval, clearer instructions or structured outputs. We reach for fine-tuning when evaluation shows a gap those can’t close.
- 01
Baseline
Measure candidate models on your evaluation set, as-is.
- 02
Ground & instruct
Add retrieval, prompt design and output schemas. Re-measure.
- 03
Adapt if needed
Fine-tune on curated examples where a measurable gap remains. Re-measure.
- 04
Route & monitor
Send each request to the cheapest model that meets the bar, and keep checking.
FAQ
Model questions.
Can we switch model providers later?
Yes. We put the model behind an internal interface and keep an evaluation suite, so switching is a configuration change followed by a re-test.
Can data stay in Australia?
We design for it: Australian cloud regions where providers offer them, or private models hosted in your Australian environment.
How much data does fine-tuning need?
It depends on the task. Quality matters more than volume; we assess what you have during scoping and advise whether fine-tuning is realistic.
Not sure which model fits?
Tell us the task and your constraints. We’ll recommend what to evaluate and how to compare it fairly.