Prism builds smart cameras that recognize people, objects and animals, and turn what they see into data. The idea is simple, but the analysis behind it isn't: machine-learning models identify and categorize elements in every frame. My job was to turn that stream of detections into answers a non-technical user could read and act on.

Role: Lead Product Designer (UX, UI and product)

Company: Prism.ai, 2018 to 2020

Team: remote design and front-end teams, working side by side with the client

Delivered: dashboard, data visualization widgets, user flows and a design system

Raw computer-vision output is noisy: thousands of detections, timestamps and video clips. Users didn't want data. They wanted answers to simple questions:

The interface also had to make setting up the AI simple: choosing cameras and defining what each one should look for.

Each "Prism" is a set of cameras with its own searches, for example Back Yard: hummingbirds, squirrels and cats. Every Prism gets a row of widgets with the latest events, proportion charts and insights. Instead of showing charts only, the system writes the patterns it finds as sentences: "Hummingbirds were most active between 6 and 8 AM", "Squirrels visit the feeder at least 4 times a day".

I designed a library of widgets for showing AI-generated data from images: event galleries, tapestry and tile views, counting, highlights, event listings, a journal, and week, month and heatmap calendars. Each Prism can be set up around what matters to that user.

A week view compares how long each element stayed in view, such as cars, bicycles and pedestrians on a road camera. Picking a day opens its video clips on the side, so users can check what the AI actually saw.

A month calendar shows the presence of a specific element across several days, so busy and quiet periods stand out at a glance. The header sums it up in one line, like "Thursday, 6th, was the most crowded day this month".

We also explored an auto-generated feed where important events are reported as they happen. The Prism Assistant joins the feed too: when it spots a new pattern in the footage, it suggests adding it to the user's searches.

I led the design across UX, UI and product, and coordinated remote design and front-end teams. A detailed design system documented every component and state for developers. The process was participatory: the client, designers and developers shaped the product together through remote design sprints, client interviews and user testing. This co-creation method is the one I later described in my book COUX, and it's why the interfaces were both easy to use and realistic to build.