← Archive

Major · Computer vision programme · NDA

Material discovery where adoption beat the model

A six-figure material library and a design team that had looked things up by hand for years. Model accuracy was necessary. Habit change was the real programme.

Lead outcome

3 hrs → 3 sec

discovery time we had to beat

Focus

Computer VisionAdoptionNDA
StackYOLOv8CLIPOCROpenSearch

What I owned

I owned the delivery runway: scope gates, stakeholder sequencing, and the definition of done that forced the tool to beat old habits — not just beat a benchmark.

Narrative

Under NDA for a global luxury fashion house. The engineering problem was indexing and retrieval. The programme problem was convincing craftspeople that a search result was safer than muscle memory.

My contribution sat in the seams: sequencing demos against real workflows, holding the bar that 'done' meant adopted — not just deployed — and keeping legal/commercial constraints visible in every sprint gate.

I do not claim the architecture. I claim the runway that let architecture land inside an organization that could not afford a clever tool nobody opened twice.

Adoption criteria in the definition of done

A model that wins a bench and loses the design floor is a failed programme.

Protect the craft, measure the minutes

Hours-to-seconds was the shared language between design leadership and engineering.

3 hrs → 3 sec

discovery time we had to beat

95%

detection accuracy at handoff

100K+

materials under the index

Next step

Want the confidential layer — names, constraints, steering artefacts? That conversation stays over email.

Email me about this case