About me

I design the intelligence layer of agentic AI: the context engineering, system instructions, and evals that decide how a model behaves. I work on both sides of the interface, from the instructions underneath to the product people touch. And I ship my own changes in code.

Professional background

At Meta Reality Labs, I led model design for agentic multimodal AI, owning both the design vision and the intelligence layer of a 20+ agent pipeline. I defined what the system needed to produce, then designed how the agents were instructed, evaluated, and improved to deliver it.

My career in model design grew out of leading design for Meta's first consumer and creator-facing generative AI products, where I owned the full product experience and the model layer underneath it. The work I'm proudest of from those years is Meta Horizon's first AI quality framework, which gave teams a shared definition of "good." That framework is now the standard across three generative AI products in Meta Reality Labs.

Before my AI work at Meta, I spent several years designing complex, high-stakes products. I worked on developer-facing API platforms, integrity systems that protect against fraud at scale, and enforcement standards built on a schema teams still use. Earlier, I designed digital experiences for mission-driven organizations including Seattle Children's Hospital and AdoptUSKids.

My path to this work came through language and design. I started as a newspaper reporter and have been designing with language since 2007, when writing for a screen was barely considered a craft, let alone a discipline. Every few years the medium of designing with language changes, but the systems thinking underneath it doesn't. I've spent most of my career building practices before the industry had a name for them, and model design is the one I'm building now.

Areas of expertise

Context engineering & prompt design

Shaping model behavior through system prompts, context structuring, and instruction design.

Evals & quality hillclimbing

Measuring technical and experiential quality, then diagnosing and fixing failures to raise pass rates.

Model behavior & agent orchestration

Defining how multi-agent systems reason, hand off work, and recover when they don't.

Product & UX design

Designing the model layer and interface together, so model responses and UI work in concert.

Language & semantic systems

Encoding editorial judgment with controlled vocabularies to make language machine-readable.

Technical fluency & partnership

Working in Claude Code to ship changes as a peer to research scientists and engineers.