Bridging data, ML, engineering, and business to build better AI solutions.
Combining modern AI/ML expertise with a decade of enterprise software engineering experience across retail, publishing, education, and telecommunications.
MouldDetect
Predicts mould risk in the home from environmental data — a seven-day risk outlook, the humidity trend behind it, and the hours worth opening a window.
APRA Q&A Agent
Answers dense banking regulation only from the official standards, always citing the exact clause — and says "I don't know" instead of guessing.
Raven AI Stylist
A conversational shopping stylist giving personalised outfit advice with AI-generated virtual try-on images.
SecondSight
A phone assistant that warns blind and low-vision users of ground-level hazards in real time — with an on-device demo video.
01MouldDetect
A data platform that helps homeowners see and manage mould risk before it takes hold — "see what your home is growing." Mould doesn't wait for scheduled inspections, so instead of reacting to visible damage, MouldDetect reads the everyday conditions that let mould grow and turns them into a plain-language outlook: the seven-day risk ahead, the humidity trend behind it, and the hours worth opening a window.
- 01
Reads the conditions that grow mould
It pulls local environmental data — humidity, temperature and weather for your area — as the signal for how favourable conditions are to mould right now.
- 02
Forecasts the week ahead
Predictive models turn that data into a seven-day mould-risk outlook, so you see what's coming rather than a single moment-in-time reading.
- 03
Explains the "why"
Each forecast surfaces the humidity trend behind it, so the risk reads as something you can understand and act on — not a black-box score.
- 04
Tells you what to do
It turns the outlook into simple guidance — like the hours worth opening a window — to keep the home dry before mould gets a foothold.
02APRA Q&A Agent
A question-answering assistant for dense banking regulation. In a regulated field, confidently answering outside the evidence is the failure that matters — so the interesting part isn't retrieval, it's the refusal. Ask about Australian prudential (APRA) standards and it answers only from them, citing the source and clause; ask something plausible but outside the corpus and it declines instead of guessing. That verifiable grounding is the product.
- 01
Answers only from the standards
It retrieves the relevant standards and generates the answer from them — never from the model's own memory.
- 02
Cites source & clause
Every answer points to the exact standard and clause, so a reader can verify it rather than take it on trust.
- 03
Refuses when unsure
Questions outside the corpus get an honest "I don't know" instead of a confident, ungrounded guess.
- 04
Eval-gated deploys
Grounding behaviour is measured on every release — if refusal regresses, deployment is blocked automatically.
- 05
Deployed to be trusted
Serverless on AWS, cost-capped and observable, with a streaming chat UI that shows the sources beside each answer.
03Raven AI Stylist
A conversational shopping stylist that gives personalised outfit advice and shows how a look might appear with AI-rendered try-on images — not just descriptions. It shows that a model trained for one domain, wired into a well-designed multi-agent system, can make recommendations feel genuinely personal: a general model runs the conversation while a fashion-specialist model handles the expert advice.
- 01
A model fine-tuned for fashion
An open base model is fine-tuned into a fashion specialist that gives sharper, more relevant style guidance than a general model alone.
- 02
Orchestrator + specialist agents
A general model runs the conversation and calls specialist agents for expert styling — the base model routes, the fashion adapter specialises.
- 03
AI virtual try-on
A dedicated agent renders a photorealistic image of the suggested outfit, so you can see the look rather than imagine it.
- 04
Streaming in the cloud
The model is served on a cloud endpoint and replies stream token-by-token for a real, responsive chat feel.
04SecondSight
A phone-based assistant that helps blind and low-vision people avoid everyday hazards — spotting obstacles on the ground in real time and warning by sound or a buzz on the phone or Apple Watch, plus on-demand scene description. It complements a cane or guide dog as a close-range safety layer, and the whole loop runs on the device, so it keeps working with no internet.
- 01
On-device hazard detection
A compact vision model runs on the iPhone and flags hazards — rocks, poles, potholes, stairs — in under a third of a second.
- 02
Automatic mode switching
The gyroscope reads camera angle: pointed down it watches the path; lifted, it switches to describing the scene — no buttons.
- 03
Offline scene description
An on-device visual language model speaks descriptions of the surroundings on request, fully offline for privacy and reliability.
- 04
Hands-free wearable alerts
An Apple Watch companion delivers haptic and speech alerts on the wrist — it can even be clipped to a cane.
- 05
Trained by an MLOps pipeline
The detection model comes from an end-to-end training pipeline and is served to the app through a small inference service.