House Points and Pip, the AI Costume Judge

D-PREP's four houses (Yellow Phoenixes, Red Dragons, Green Griffins and Blue Unicorns) face off every year in Theme Week. In 2025 I built a live scoreboard for it. It did the job for one week, but anyone with the shared password could change a score, and nothing recorded who gave points or why.

The goal for 2026 was bigger: a house points system the school could use all year, for good work in class, assembly games and future events, with Theme Week (September 14 to 17) as its first event. It had to serve about 250 students and 34 staff, earn their trust, and run at essentially no cost. It now lives at dprepstuco.org, in place of the old site.

Most of the people giving points are teachers between lessons. So every role opens on one obvious action, the labels are plain, and a receipt confirms the student, the house, the points and the reason.

Sign-in is a code sent to the school email address. I tested Microsoft sign-in first, but it needed approval from the school's IT administrators, so I switched to email codes rather than wait.

Teachers and student leaders give points with a written reason, and every award records who gave it. Mistakes are corrected with a new entry, never erased, so the history always adds up. Admins can add one student or import a whole list, with a preview before anything changes, and the standings are public, with no login needed to follow along.

Judging costumes for a whole school, four themed days in a row (Survival, Mystery, Fantasy and Adventure), is slow and hard to keep fair. So I built Pip, a costume judge that lives on a kiosk: an iMac with a camera and microphone, and a supervisor nearby on an iPad.

A student says "Hey Pip", tells Pip who they are and strikes a pose. Pip takes the photo, a vision model grades the costume against the day's theme, and the points go straight to the student's house. A top score gets a happy dance and confetti.

Pip listens and speaks on the machine itself, so nothing said while it waits leaves the room. Each person can enter once a day, however many times they try.

An AI judge is only useful if students trust it. Pip grades on a four-step rubric, from no visible theme detail up to a distinctive interpretation, and rewards what it can see, never cost or appearance. School uniform alone earns zero.

When a photo is unclear or the judgment is uncertain, Pip stops and hands the decision to a person. Supervisors can correct any score, and every correction is logged. Before the week I tested several vision models side by side, including photos with no costume in them and notes trying to talk the judge into full marks.

Privacy was a design decision from the start. Costume photos stay in a private catalog only administrators can open, and students' nicknames are never sent to the AI provider.

The week closed with house performances and an audience vote. Students scanned a QR code and voted on their phones in English, Thai or Simplified Chinese, while the projector showed live ranked totals that reorder as the points change.

One Hide points switch can blur every score on every screen for a reveal, without touching the data underneath.

Over the four days Pip judged around 200 costumes a day, roughly 800 in all, and the finale drew 430 audience votes. The Red Dragons took the week.

Theme Week was only the first event. The same system now carries house points through the rest of the school year.