Optimal Dynamics · Tech lead, frontend · 2021–25
Dispatching: AI recommendations people can trust
The flagship product of an AI decision platform for freight logistics. I led its frontend from concept to v1 in four months — the interfaces where dispatchers act on, question and override machine recommendations.
00
What I did
Wrote user stories and ran sprint planning and retros as team second-in-command.
Human-in-the-loop flows: when to trust, question or override the model.
Live-updating UI, feature flags, canaries and kill-switches; component library foundation.
01
The problem
Freight dispatchers make dozens of high-stakes assignment decisions a shift. An ML model could recommend better ones — but only if people understood and trusted it enough to act.
02
Recommendations that move
Recommendations refresh on 1–5 minute cycles. The UI updates in place without losing a dispatcher's place or selection, and makes it obvious what changed.
03
Explain the why
LLM-powered explanations sit beside each recommendation, so dispatchers can see the reasoning before committing a truck.
04
Overrides that teach
When a dispatcher chooses differently, the override flow records why — feeding the model team real signal instead of silent disagreement.
05
Shipping safely
Mission-critical tooling can't go down mid-shift. Feature flags (LaunchDarkly), canary deployments and instant kill-switches kept uptime at 99.99%+ while we shipped weekly.