Latest visual demo · Curling
Computer vision that tells a story.
From house geometry and stone positions to a tactical explanation. Watch the recorded sequence, then inspect the evidence behind it.
July–September 2026
New evidence to explore.
New case studies in computer vision, quantitative research and the work of leading dependable data products.
What the snooker model gets wrong
A frozen snooker model comparison reveals fewer misses, persistent extra detections and the importance of reviewing the labels.
One control plane. Several sports.
Shared interfaces, observable pipelines and clear ownership across independently operated sports-data systems.
A dashboard should earn your trust
Building reporting people can rely on: consistent cohorts, complete records, visible uncertainty and a dependable read path.
Earlier demo · Snooker vision
Computer vision: from frame to story.
The 56-second sequence moves from a frozen table state through ball labels, shot geometry, temporal updates and an end-user overlay.
Read the build storyDemo library
Tools and workflows you can inspect.
Ship drop simulator
A client-side probability lab for drop rates, expected value, pity rules and collection odds.
Attimo poker workbench
Four approved interface views covering session overview, beats, arcs and editorial variants.
Lacuna reconstruction workbench
Inspect retrieved evidence, candidate readings and uncertainty in a responsive source-first interface.
Invoice_gen
An installable desktop app built around a real small-business workflow gap.
Information-boundary walkthrough
See how chronological folds and zero odds inputs keep the model evaluation honest.
Live-shadow evidence pack
Public aggregate charts show scale, horizons and promotion states without exposing live commercial signals.
Chronological model audit
Inspect the research surface, forward-moving windows and paired-orientation leakage controls.
Data-product operating model
A visual route from signal through evaluation, ownership, release and post-launch learning.
Have a demo-shaped problem?
Make the risk inspectable.
A focused prototype can align a team faster than another architecture deck.