Financial & regulatory text
Filings, disclosures, contracts and compliance documents.
AI Consulting
Build the thing, and prove it works.
Most AI projects fail somewhere between a promising demo and a system anyone trusts in production. We work on that gap: scoping the problem properly, building the system, and measuring it against a benchmark you agreed to before we started.
Engagements
Three shapes of work.
2–3 weeks
Scoping sprint
We take a workflow you think AI should handle and come back with an honest answer: what's feasible now, what isn't, what it would cost, and how you'd know it was working. Sometimes the answer is don't build it.
6–12 weeks
Build engagement
A working system delivered into your environment — retrieval, agents, evaluation harness and the operational scaffolding around it. Your team is in the codebase from week one.
Ongoing
Evaluation & review
For teams who already built something and can't tell whether it's good. We construct the labelled set, run the comparison, and report where it fails.
How we work
Evaluation first, model second.
The first deliverable in every engagement is the measurement: what does good look like, who labels it, and what's the baseline to beat. Without that, a system can only be assessed by whether the demo felt impressive.
This is the same discipline we apply to our own products — the benchmark on our research page is empty until the numbers exist, and we hold client work to the same standard.
- Fixed scope and fixed price on scoping sprints
- Your team owns the code and the evaluation harness
- No lock-in to our models or our infrastructure
- We'll tell you when the answer is a spreadsheet, not a model
- Based in Sydney; comfortable working across AU, NZ and US hours
Domains
Where we're strongest.
Multi-agent systems
Specialised readers whose outputs need reconciling into one decision.
Retrieval with citations
Systems where an unsourced answer is worse than no answer.
Evaluation design
Turning "it seems better" into a number someone can contest.