Founding Engineer
Video reasoning, retrieval over production data, and the canvas itself
The product is a real-time collaborative canvas with agents underneath it, and the hard parts are all still open. Yjs sync across a canvas holding well over a thousand items. An agent loop that owns prompting, model selection and reference search without taking direction away from the filmmaker. Retrieval over two million film stills.
You'd pick one of those and own it, then be responsible for it in production, because the studio next to you is shipping client work on it this week and you'll hear about it either way.
What you'd do
- Own a hard surface end to end: the agent loop, the sync layer, or retrieval.
- Ship to a canvas that real productions are using, with real deadlines attached.
- Work directly with the filmmakers using it: the feedback loop is a room, not a ticket queue.
- Make the calls on architecture in your area, and live with them.
What we're looking for
- You've built something hard and shipped it to people who noticed.
- Depth in one of: real-time collaboration and CRDTs, agent systems, or retrieval and video models.
- You care how software feels, not only whether it works.
- You want an owner's surface, not a well-specified task.
Every role is full-time, in person, and based in Los Angeles. Four questions, and the only one that matters is the link to your work: work beats a cover letter.