Content workflow UX
The Worthington pipeline: a workflow humans and machines both navigate cleanly.
Built a content production system where AI generates, a human reviews, and the system auto-publishes only after an explicit decision. The UX challenge was making the hold-period approval flow feel natural rather than bureaucratic — for a non-technical client managing their own site.
- Admin approval UI: immediate publish, reject with feedback, or let the 3-day hold auto-expire to publish
- Weekly market research toggle — enables OpenAI web-search topic augmentation without complexity for the operator
- Pricing matrix admin panel: client-configurable guide prices that AI-generated cost posts are constrained to use
- Post management: publish/draft toggles, slug editing, image upload, manual content override — all without touching code
- Image modes (AI-generated / upload / branded fallback) exposed as a simple admin radio, not a config file
Desktop tool UX
BAM Band Removal Engine: a guided operator workflow for a technically hard problem.
LED/PWM banding in arena sports photography is a real, frustrating problem with no off-the-shelf fix. I built a PyQt6 desktop app with a guided 4-step calibration flow — designed so a photographer can produce venue-specific correction profiles without understanding FFT algorithms.
- Step bar guides the operator: Open image → Paint background → Analyse → Save profile
- Interactive mask painting tool for defining reference regions — the UX decision that makes the maths usable
- Heatmap overlay, banding strength gauge, estimated band period, and row-profile chart — all in the same view
- Before/after split compare with draggable split line, zoom/pan, multiple view modes
- Venue profile system: save settings per arena, reload next time, batch-process a full event shoot
Systems thinking
Clinical diagnostics: where UX decisions have patient-level consequences.
Working as Head of Commercial & IT in cellular pathology meant evaluating digital pathology platform UX (Indica Labs HALO, Deciphex, 3DHISTECH) against real-world adoption by pathologists and BMS staff who had zero tolerance for friction in a diagnostic workflow.
- Evaluated and procured pathology AI platforms (Paige AI, Indica Labs) — UX and workflow integration were procurement criteria alongside accuracy
- HL7 integration work required mapping clinical data flows to system interfaces that non-technical lab staff could use reliably
- ISO 15189-adjacent validated environments: every interface change required documented rationale and formal validation evidence
- The instinct developed: in high-consequence environments, operator error is a design failure, not a training failure