> lab · EXPERIMENTAL BUILD INDEX
Prototypes, internal tools, and technical tests. Deliberately unfinished — this is where ideas get proven or abandoned before they go anywhere near a client project.
READ THIS FIRST
Nothing here is a failed client project. Client work is in the Work index and all of it is live.
The Lab is the opposite: no deadline, no client, and permission to be wrong. Some of it ends up shipping.
A narrow AI assistant trained only on one business’s own documents — pricing, services, policies — and hard-limited to answering from them.
The interesting problem is not the model. It is refusing to answer confidently when the answer is not in the source material.
Turns a photo of a vehicle and three fields into a draft service quote for review. Never sends anything to a customer on its own.
Accurate enough to save time on the common cases, wrong often enough that a person still signs off. That balance is the whole design.
Culls a photo shoot down to a shortlist of selects — sharpness, expression, duplicates.
Right roughly seven times in ten, which is not good enough to trust and not bad enough to abandon. Currently parked.
A tablet display for the studio floor: today’s jobs, current status, next drop-off. No login, no interaction, always on.
Running at Axiom. The lesson was subtraction — every feature added made it worse as a glanceable display.
Classifies an inbound enquiry by urgency and service type from free text, before a human reads it.
Grew out of the Comfort Homes 24 symptom triage. Generalising it without making it vague is the open question.
The device-switching live site preview used on this website’s work modules — static capture, live embed, desktop and mobile.
Started here as an experiment in showing real work instead of screenshots. It is now part of the site you are reading.
Client work has to be right on the first try. The Lab is where the second and third tries happen — cheaply, privately, and before anyone is paying for the outcome.
If an experiment survives real use inside my own businesses, it becomes something I will build for someone else. Most do not survive, which is the point.
Longer write-ups on AI, software, design, and running a business are planned. Not published yet — when there is something worth reading it will appear here.
NEXT
Several of these started as a client asking whether something was possible. That is a good way to start a conversation.