Led design for Humanloop, a platform that helps developers build, evaluate, and ship applications powered by large language models. Over four years as first designer and Head of Design, I designed the brand, the product through three redesigns, and the design system underneath it.
The product changed shape three times as the industry moved beneath it — from ML annotation tools to prompt management to a full developer workspace. Each redesign introduced a new organising concept. The last treated prompts, datasets, evaluators, and tools as versioned files that developers could branch, compare, and ship like source code. That structure is the one that held.
It made evaluations natural. A developer runs a prompt against a dataset, scores the output with an evaluator, and sees exactly where a new version regresses — all versioned artefacts composing in the same workspace. The same model extended to agent workflows: multi-step pipelines of prompts and tools, tested the same way a single prompt can be. Particular care went into making each surface feel like an extension of how developers already work with code — not a dashboard layered on top of it.
Enterprise teams at Vanta, Gusto, and Duolingo used the platform. In 2025, Anthropic acquired Humanloop.
