Sitters

Context

At a museum you read a forty-word label, glance at the painting and move on. The people in the portrait can’t tell you anything.

At the Anthropic Hackathon in September 2026, I asked a simple question: what if you could talk to the person in the painting, to the painter, or to both?

What we built

My two teammates and I built Sitters, and I implemented the app myself. You point a camera at a painting and hold a voice conversation with its sitter, its painter, or both.

The idea. Talk to the person in a painting, to the painter, or to both.
The spec. Written before any code, with the rules below built in.
The build. A Next.js app with Claude and ElevenLabs, which I implemented.
The personas. Each one separates sourced answers from labeled speculation, with a curator layer for context.

Code to comeThe code behind Sitters, the museum app.A link to its repository, with keys and private data removed.

The decision

Before any code existed, we wrote the rules into the build spec: no living people, no knowledge beyond the sitter’s lifetime, and anything not from the historical record would be marked as imagined.

I designed the conversational personas around those rules: each one separates sourced answers from labeled speculation, and a curator layer adds the context.