Taste Labs

Building the "taste layer" for AI: preference datasets, rubrics, and eval environments for frontier labs, powered by a vetted network of expert tastemakers. $18.5M seed from Amplify and CRV.

AI Training AI Training Data / Preference Data / Expert Network
Visit Taste Labs
★ ★ ★ ★ ★ New (0 reviews)
Remote-First (Global tastemaker network) HQ
20 employees
2025 founded

What is Taste Labs?

Taste Labs is building what it calls "the taste layer for AI": the data and infrastructure that gives AI models and agents judgment about subjective quality. On one side it supplies frontier AI labs with preference datasets, rubrics, and evaluation environments designed to improve how good a model's output actually is, not just whether it is factually correct. On the other it gives application-layer companies the context and verification tools their agents need to produce more on-brand and genuinely creative work. The thing that makes this possible is human: Taste Labs curates a vetted, trust-based network of experts and "tastemakers" who supply the high-quality subjective judgment needed to train and evaluate models on taste-dependent tasks. Its initial focus area is design. The company launched out of stealth with an explicitly anti-"AI slop" thesis, arguing that as AI output volume explodes, judgment and quality become the scarce inputs. Taste Labs was founded by Thais Castello Branco, previously growth lead at Exa, and has grown to a team of around 20, including a founding designer who spent 15 years as a VP of Design and acts as the internal arbiter of quality, plus founding engineers drawn from Exa, Palantir, Mercado Libre, and research institutions. It raised an $18.5M seed co-led by Amplify Partners and CRV, after taking its first cheque from Latitud, the pre-seed firm backing Latin American founders.

Mission & values

Give AI models and agents taste, by building the preference data, rubrics, and evaluation infrastructure (backed by real human experts) that make subjective quality measurable and trainable.

Qualifications

Two paths. (1) The tastemaker / expert network: Taste Labs recruits vetted experts to provide the human judgment behind its preference data, rubrics, and evaluations. The initial focus is design, so working designers with genuine craft and a strong portfolio are the core profile, with the network expected to expand into other taste-dependent domains over time. Entry is trust-based and curated rather than open sign-up, so expect real portfolio vetting rather than a quick signup form. (2) Full-time roles at a ~20-person, well-funded seed-stage startup: engineering (ML, data infrastructure, backend, full-stack), applied research on evaluation and preference modeling, design, product, and operations for running the expert network. Application flow lives at tastelabs.com and its LinkedIn page. Because the company is early and remote-first, hires are highly leveraged and equity is a meaningful part of compensation. Experience with RLHF, preference data, evals, or building and managing expert/creator networks is directly relevant.

Leadership

T

Thais Castello Branco

Founder

Founded Taste Labs. Previously growth lead at Exa. Built the company around the thesis that as AI output volume explodes, human judgment and taste become the scarce, valuable inputs.

Hiring process

  1. 1

    Apply at tastelabs.com or via LinkedIn

    Two paths: apply to join the vetted tastemaker / expert network, or to an open full-time role. Both start on the site or its LinkedIn page.

  2. 2

    Portfolio or profile vetting

    Expert network: real portfolio and craft review, since entry is curated and trust-based rather than open sign-up. Full-time: recruiter or founder screen.

    About 7 days

  3. 3

    Calibration or technical deep-dive

    Experts: a calibration exercise to check judgment against the platform's quality bar. Full-time engineering: coding plus discussion of evals, preference data, or data infrastructure.

    About 7 days

  4. 4

    Team interviews

    Conversations with the small founding team on collaboration, standards, and depth in your area.

    About 10 days

  5. 5

    Onboarding or offer

    Experts: onboarding to the rubric and task workflow. Full-time: offer with base plus meaningful early equity.

    About 5 days

Funding

StageSeed
Total raised$18,500,000
Investors
Amplify Partners CRV Latitud

Awards & recognition

  • $18.5M seed co-led by Amplify Partners and CRV · 2026

    Taste Labs

  • Launched out of stealth on an anti-"AI slop" thesis · 2026

    Taste Labs

Company information

  • Industry: AI Training Data / Preference Data / Expert Network
  • Location: Remote-First (Global tastemaker network)
  • Founded: 2025
  • Employees: 20
  • Website: tastelabs.com ↗
  • LinkedIn: Company page ↗

Frequently asked questions

What does Taste Labs do?
It builds the "taste layer" for AI: preference datasets, rubrics, and evaluation environments that help frontier AI labs improve the subjective quality of their models, plus context and verification tools so application-layer agents produce more on-brand, creative output.
What is a "tastemaker" at Taste Labs?
A vetted expert in the company's curated network who supplies the human judgment used to train and evaluate models on subjective tasks. The initial focus domain is design, with expansion expected into other taste-dependent areas.
Can anyone join the expert network?
No. Entry is trust-based and curated rather than open sign-up, so expect genuine portfolio and craft vetting. Working designers with strong craft are the core profile today.
Who founded Taste Labs, and who backs it?
Thais Castello Branco, previously growth lead at Exa. It raised an $18.5M seed co-led by Amplify Partners and CRV, after a first cheque from Latitud. The team is around 20 people.
How does this differ from regular data annotation?
Most annotation work targets objective correctness. Taste Labs targets subjective quality: whether output is actually good, on-brand, and well-crafted. That requires vetted domain experts rather than a general crowd, and the work is closer to expert review than commodity labeling.
How much does it pay?
Not publicly published. Expert-network pay varies by engagement and domain and typically sits well above commodity annotation for this kind of expert review. Full-time roles follow well-funded seed-stage bands with meaningful equity. Confirm per role.

Stay in the loop.

One email per week, 5 hand-picked roles.