I'm a Sydney mortgage broker. I rebuilt my own admin — email, documents, follow-ups — around AI agents that run whether or not I remember to ask. Now I build the same for other small operators.
Document set → pull the numbers that matter
WaitingHandledInbox → sorted, replies drafted
WaitingHandledExpiring Thursday → follow-up drafted
WaitingHandledApplication form → pre-filled from file
WaitingHandledWeekly status → compiled and filed
WaitingHandledSorted and drafted by 9am. I still decide what actually goes out.
Every one of these started as something I needed for my own business first.
Assistants that take over the repetitive parts of running a business — sorting and drafting email replies, pulling data out of documents, pre-filling forms, chasing the follow-ups that would otherwise slip. Built around your actual workflow, not a generic chatbot.
Checks that run on schedule without being asked. Lead handling that doesn't drop anyone. Spreadsheet work that stops being manual. The measure of success is how much less you have to hold in your head.
Fast, static, no platform lock-in and no monthly fees. Built to load quickly and turn up in local search. Ready-made if you want to be live today, custom if you need something specific.
Most "AI help" means opening a chat window and asking a question. This is different in kind: work arrives, gets routed, gets remembered, and gets prepared on a schedule — with a person approving anything that reaches a customer.
A message, a note, a forwarded email — from wherever I happen to be. It gets routed to whichever specialist handles that kind of job. No app to open, no dashboard to remember.
Everything lands in a long-term knowledge base. Context accumulates instead of starting from zero every conversation, so the answer in March knows what happened in January.
Inbox sweeps, follow-up checks, project status — running whether or not I think to trigger them. The work that gets forgotten is the work that costs you clients.
Source-of-truth documents, phased delivery, independent code review before anything ships. The failure mode I care about isn't "it gave a wrong answer" — it's "it broke quietly three weeks ago and nobody noticed."
Not case studies for clients I can't name. These are mine — one live and public, one running at home every day.
The website for my mortgage broking practice. Static build for speed, with accessibility, performance and HTML validation checks running automatically on every change — the same discipline I'd apply to a client's site, applied to my own first.
Visit the siteAn AI learning companion I built for my two sons. Multi-subject tutoring that remembers what they've already covered, reading support, voice conversation, and a dashboard so I can see how they're going. It isn't a product for sale — it's a real one, in use at home, and the hardest thing I've built.
Read how it's builtYou tell me one thing you do every week that you resent doing.
I map out what an assistant for it would actually look like — free, and yours to keep either way.
I build the small version first. You use it for a fortnight before we talk about anything bigger.
The one that eats an hour every week and always lands at the worst time. I'll tell you honestly whether it's worth automating — sometimes the answer is no, and that's a useful answer too.
Replies come back the same business day. Prefer email? Write to me directly.
Eden is the SunVex AI front desk. She can answer the basics, collect the right details, and route important enquiries to the SunVex team.
Eden is an AI assistant, not a person. Pricing, commitments and sensitive enquiries are reviewed by the SunVex team before anything is confirmed.