Aurora CPA — an AI-native accounting platform
Region-pinned Australian inference with evidenced data residency
Accounting firms hold some of the most sensitive records a business produces, under professional obligations that do not bend because a tool is convenient. Aurora CPA wanted the working advantages of applied AI — faster answers out of their own documents, a front door that could scope an engagement without a phone call — without moving client data somewhere they could not account for it.
The Challenge
The two goals pull against each other. General-purpose assistants are useful precisely because they are broad, and broad is the wrong property for a firm that must be able to say where a given document was processed and who saw it. The constraint was not "add AI"; it was to make each capability provably bounded.
The Solution
Retrieval runs over Aurora’s own Microsoft 365 corpus under per-tenant row-level security, so the isolation between clients is enforced by the data layer rather than by prompt instructions asking a model to behave. Inference is pinned to Azure OpenAI in the australiaeast region, and residency is evidenced rather than asserted. Every consequential action writes to a hash-chained, append-only audit log held under seven-year WORM retention, so the record cannot be quietly revised after the fact.
The public site is pre-rendered and structured for answer engines as well as for readers, and the quote assistant in front of it runs a frozen system prompt behind a Turnstile challenge — a narrow, fixed surface instead of an open chat box. The whole environment is provisioned as infrastructure-as-code, so the residency and retention guarantees are properties of a reviewable definition, not of one correctly-clicked console.
Key Technologies
- Inference: Azure OpenAI, region-pinned to australiaeast, with evidenced residency.
- Retrieval: RAG over a Microsoft 365 corpus with per-tenant row-level security.
- Auditability: hash-chained append-only logging, seven-year WORM retention.
- Front end: pre-rendered, machine-readable marketing site with a scoped quote assistant.
- Infrastructure: fully IaC-provisioned Azure.
Outcomes
Aurora runs applied AI over its own records without the data leaving Australian infrastructure, and can demonstrate that rather than assert it. The audit chain means an assurance question has an answer that survives inspection — which, in a regulated profession, is the difference between a tool that can be adopted and one that cannot.
Aurora CPA is at auroracpa.com.au, published with their permission.