← All organizations

AI training data and design evaluation infrastructure

Taste Labs

Taste Labs builds data and infrastructure for teaching AI models and agents to assess subjective quality, starting with design. It sells preference datasets, reasoning data, rubrics and evaluation environments to frontier AI labs. For coding agents and creative applications, it provides context, evaluation and verification infrastructure to help generate designs that reflect users’ preferences and brand requirements. Its focus is on turning aesthetic judgment into signals that models can learn from and tools that applications can use.

Founder and CEO Thais Castello Branco previously led growth at Exa. Taste’s approach centers on a curated evaluator community: people vetted for aesthetic judgment nominate others, while software helps them produce more consistent assessments. The aim is to teach models to recognize poor outputs before adapting to individual tastes. Its exploratory research includes agents that reconstruct website pages to match a brand; observed reward hacking highlights the difficulty of evaluating visual work while allowing creative freedom.

Taste Labs emerged from stealth in 2026 with $18.5 million in seed funding, co-led by Amplify Partners and CRV. At its June 2026 launch, Taste had a team of 20 and was already working with frontier labs and application companies.

tastelabs.com

1 talk

Newest first

1 speaker at AIE

Affiliations reflect their AIE appearances, not necessarily current employment.

Messages from the stage

Evaluation without flattening taste

The talk examines limitations of LLM-as-a-judge and reward hacking, arguing for expert-curated preference data and careful human quality assurance while preserving legitimate differences in taste.

Affiliations reflect each recorded session, not necessarily current employment.

Company sources · checked 2026-08-28