AI, whatever you need.

If you can describe the problem and what a good outcome looks like, we can scope it. The Audit is where we agree the shape, the model, the review surface, the data path, and the price. We have scoped document work, drafting workflows, classification pipelines, retrieval systems, reconciliation tools, and agents with bounded authority. Your brief may be none of those, or a combination we have never seen before. The engagement works the same either way.

Yes. Where the brief asks for an agent, we build an agent. What we design into the workflow is a human sign-off on anything irreversible. Not because we refuse to automate those steps, but because a regulator, an aggregator, or a PI insurer is going to look at that output sooner or later and ask who authorised it. Agents we have scoped have had narrow documented permissions and a named operator on any action that cannot be undone.

Chosen per engagement, documented in your runbook. Most document and drafting work runs on frontier models (OpenAI, Anthropic, Google) via API with zero-retention settings. Smaller extraction or classification tasks often run on smaller open models hosted in an Australian region. The data path is specified in writing for every build: what is sent, to which endpoint, what is retained, for how long, and what the rollback looks like if the provider changes terms.

Review catches it. Outputs with client, regulator, or financial consequence pass through a human checkpoint before they leave the system. Low-confidence outputs flag themselves, escalate, or fall back to manual handling. When a provider updates a model and behaviour drifts, monitoring triggers a rollback to the last known-good configuration while we investigate. The runbook documents every step of that process.

SaaS AI products are built for the median user. A compliance-bound Australian practice is not the median user. A bespoke build costs more up front and less over time, because you are not paying a monthly subscription for features you do not use, working around ones you cannot change, or exposed to a vendor pivot that breaks your workflow. You own the prompts, the model configuration, the credentials, and the documentation from day one of handover.

Occasionally, if the scope is genuinely small and sits inside a platform we know well. In most cases the two-week Audit is worth it. AI work fails most often because the scope was wrong, not because the model was wrong. The Audit is where we decide what to build, which model suits the job, what the review surface looks like, and what the rollback path is. That work gets done once. Doing it up front is cheaper than doing it mid-build.