A case and proposal platform where sensitive data never leaves one AWS account.
StudyBird Australia is an international education agency. Feneco is building the platform its counsellors use to manage student cases and proposals — course selection, fees, scholarship terms, visa and medical documentation — with AI embedded where it earns its place, and kept out where it doesn't.
Data that can't leave the building
An education agency's case files carry some of the most sensitive data a business can hold — passport details, visa history, medical records. Sending that to a public model API is either a compliance problem or a commercial one. So the AI in this build runs on open-weight models hosted entirely inside StudyBird's own AWS account in Sydney. Nothing in a case file touches a third-party model provider — it never leaves the account it was uploaded to.
The problem with asking a model to remember a fee
The brief asked for an assistant that would hold course fees and scholarship terms. That sounds like a natural fit for an LLM — until you test what happens when it's asked for a figure it doesn't have. Language models don't store facts reliably; they produce a convincing answer whether or not the underlying fact exists. In a proposal platform, a quote with an invented fee isn't a UX quirk — it's a real liability, commercial and reputational.
The design puts a structured database in front of the model: the model retrieves and writes against that database, and it never generates a number itself. If a fee is missing, the interface shows a blank with a warning — not a plausible-looking estimate. The model handles language; the database handles facts.
The shape of the build
Open-weight models on infrastructure you own, in the region data must stay in
Human approval modelled as a workflow state, not promised in a disclaimer
An audit trail on every AI-assisted decision
Phased delivery — the first stages ship value with no AI at all
Status
StudyBird is currently in development. The first delivery stages are shipping without any AI involved at all — the case and proposal workflow itself — so the agency gets real value before the highest-risk, highest-reward part of the build lands.
If your team is weighing where AI can — and can't — touch sensitive data, let's talk.