Education
Master of Data Science, UWA. Bachelor of Commerce, Finance, Curtin.
Matthew Haskins · Perth, Western Australia
Matthew Haskins
Hi — I built FaceGate.
I’m a data scientist and AI engineer based in Perth. I work on applied machine-learning systems — particularly the kind that have to be reliable, explainable, and easy for non-technical people to use.
FaceGate is a side project I built because the idea kept coming up in conversations with friends who work in the childcare industry — about the trickiness of publishing school photos responsibly.
Find me
Credentials
The shape of the work I do.
Professional
AI Engineer at EY — digital-asset assurance, blockchain analytics. Prior finance and tax roles.
Research
Master's thesis on temporal Graph Neural Networks for rumour detection.
Practice
Builds full-stack tools across AI, data visualisation, and developer tooling. Cares about design, clarity, and things that work.
The story
Why I built this.
The conversation that kept coming up in 2024 was the same one. Schools and childcare centres take a lot of photos. Some of those photos contain children whose families have asked for them not to appear in newsletters, on social, or in the yearbook. The systems for tracking who’s consented to what are usually spreadsheets and inboxes. The system for actually checking before a photo goes out is usually a person, eyeballing a hundred photos at the end of a long day.
Off-the-shelf face recognition exists, but the answer it gives a school is the wrong shape: send photos of children to a vendor’s servers, and trust that vendor with biometric data. Solving a privacy problem by creating a new one.
The brand promise is the architecture. The phone is the only computer involved.
FaceGate is the version that doesn’t do that. The phone is the only computer involved. The consent list lives on the device, the matching runs on the device, the audit log stays on the device. There is no FaceGate server because there is no FaceGate server to build.
What I deliberately didn’t build is just as important: no cloud upload, no user login, no telemetry beyond the standard app-store metrics, no “social features”, no live chat support widget that watches every page.
The interesting work was making this useful for people who don’t care about the architecture. A Marketing Coordinator at a primary school shouldn’t need to think about cosine similarity. The app gives them a verdict per photo, the option to override it, and an audit trail. That’s the whole product.
What’s next
The next twelve months, honestly.
Now
WA school pilots
Working with WA schools on early deployments — watching how the workflow lands in a real comms team and tightening the rough edges that only show up in use.
Q4 2026
First cohort model card
Assessing real-world performance on the pilot deployments — measuring accuracy across the cohorts the app actually sees, and publishing the results alongside the model card.
Beyond
Adjacent verticals
Aged care, childcare integrations, ACCHO partnerships. Same architecture, same guarantees — different consent contexts.
If you’re thinking about FaceGate for your organisation, I’d rather hear about your workflow before pitching mine. Get in touch.