Computer graphics · machine learning
William Koch
I work on spatial intelligence for world models.
I’m a PhD candidate in Computer Science at Princeton University, advised by Felix Heide. My research sits between geometric learning, graphics, and learned scene representations.
Updates
Recently
- SCION accepted at NeurIPS 2026Scene composition with instanced neural primitives.↗
- Keryx releasedMy voice agent harness, previously Jarvis, is now open source.↗
- Research Memory presented at the Agentic AI SummitBerkeley, California.↗
- ScenarioControl accepted at ECCVVision-language controllable scenario generation.↗
- AI Workflow Automation talk at PIXLA skill-sharing session on research automation and agent-assisted workflows.↗
Selected work
Highlighted papers
Software & experiments
Highlighted projects
Pre-1.0Keryx, a voice agent harness
A personal voice agent you can phone. It hands your request to coding agents on your machine and calls you back when the work is done.
Public betaResearch Memory
A local-first research memory for people and agents, built on plain Markdown, hybrid search, and MCP.
In developmentreMarkable AI integration
Write @claude in a reMarkable notebook and get an answer, a chart, or a pull request back in the same notebook.
Stay in touch