Employment work

Rely

Frontend Lead & Product Designer·2025·production

Next.js, Go, WebSockets, Tailwind v4

A Hub of Stations: live Preview with an OK/KO on the piece, and a Recipe graph engineers deploy to the floor.

Station Preview with a live PCB camera feed, piece identity, defect count, and OK/KO totals — Preview is the floor: the piece, a verdict, OK/KO/Rework. The red card is the status, not a chart.

Decisions

1. The Station is the product

The Hub is a list of Stations, not a model zoo. Opening one is the floor: live camera, piece identity, OK/KO. The operator reads a verdict. Recipes, overlays, and analytics sit beside that surface — they do not replace it.

2. A Recipe is a graph you deploy

Inspection is not a script per line. Trigger → images → detect → Evaluación is a visual flow with a contextual inspector. Evaluación is a rule on detections (count, confidence) that writes a station status: OK if it holds, KO if it does not. Deploy publishes the Recipe; the canvas draft is not what the floor is running.

3. Overlays are the operator language

Detections have to be readable on the image: label corner, dashed ROI, human vs model. That overlay is configured per Station so Preview and review share the same picture language — status is the color, not the class name.

Result

In production on factory floors, multi-tenant / multi-site. Hub in the cloud, an agent per Station on the line.

Proof

Hub Stations list showing Linea 1 with a live thumbnail and an empty Linea 2 card — Hub: Stations in the org. Enter one to work the line.

Recipe flow canvas with trigger, images, detector, live view, and Evaluación inspector — Recipe as a graph: trigger → images → detect → Evaluación. Inspector on the right; Deploy publishes to the Station.

Overlays settings with label corner, dashed ROI, human vs model, and a preview of detection boxes — Overlay language: label corner, dashed ROI, human vs model. Same picture on the floor and in config.

Station analytics layout editor with KPI widgets and a catalog to add first-pass, reject, and volume panels — Analytics (proof): station layout of widgets — first-pass, reject, volume — added from a catalog, not a fixed dashboard.