Miju Labs

← All people · Taste Labs

Research-side Member of Technical Staff at Taste Labs. She wrote the company's public research agenda and runs its Prototype fellowship for outside researchers. Before Taste Labs she worked on the economics and provenance of AI training data, a rare background in this market.

high confidenceSan FranciscoAt Taste Labs since 2026Updated 2026-09-19

Background

McGill University undergraduate (computer science and mathematics, with interests in economics). Emergent Ventures grantee (a project on talented youth in Nigeria), 2023 Interact fellow, non-resident fellow at the Institute for Progress writing on high-skilled immigration, research affiliate at Topos Institute, contributing editor at Reboot, and previously a software-engineering intern at Amazon.

Her papers include 'Consent in Crisis: The Rapid Decline of the AI Data Commons' (2024, a Data Provenance Initiative study), 'Anatomy of a Machine Learning Ecosystem: 2 Million Models on Hugging Face' (2025), 'The Economics of AI Training Data: A Research Agenda' (2025) and 2026 work on agent trajectories as programs. Her GitHub tools procgrep and bdtrace study and export coding-agent traces.

What they run now

Sets and publishes Taste Labs' external research agenda. The four areas are models that design, the geometry of the visual web, human interaction as preference signal (including reward-hacking detection) and version history for design processes. She also runs the Prototype fellowship, which offers grants, compute and access to Taste Labs' human datasets to HCI, ML and design researchers.

Career

  1. 2026 – presentMember of Technical Staff, Taste Labs — first bylined posts Aug 2026
  2. 2024Non-resident fellow (high-skilled immigration), Institute for Progress
  3. —Software Development Engineer intern, Amazon
  4. —Research affiliate, Topos Institute
  5. —McGill University, Undergraduate, computer science / mathematics

On the record

Prompts are lossy · 2026-08

Natural language is lossy, so any design prompt allows many interpretations; intent has to be inferred from richer signal. source

Interaction as preference signal · 2026-08

Human interaction inside design workflows (edits, choices) should serve as preference data, to measure creativity and detect reward hacking. source

Design history · 2026-08

Creative work needs version control and traceability beyond code diffs; 'design history for agents' is an open problem. source

Recent posts

22 posts archived · most engaged first, then the latest

Hamidah Oderinwale@didaoh·
1/ With @BenDLaufer and Jon Kleinberg, we constructed the largest dataset of its kind to date: 1.86M Hugging Face models. In a new paper, we mapped how the open-source AI ecosystem evolves by tracing fine-tunes, merges, and more. Here's what we found 🧵
84122860KView on X ↗
Hamidah Oderinwale@didaoh·
Earlier this year, I saw posts about the highly coveted NSF Graduate Research Fellowship award decisions. Unfortunately, international students are ineligible, as is the case with many federally funded fellowships. For @IFP, more on why this matters and what can be done.
7199126KView on X ↗
Hamidah Oderinwale@didaoh·
Super excited to see the fellowship out in the world! Come work with us at @tastelabs on some exciting problems (or pitch us on better ones): from fingerprinting model generations, modeling aesthetic human preference, to building the post-training stack for the unverifiable. DMs open :)
1258212KView on X ↗
All 22 in the feed →

Connections

Who they amplify

Accounts whose posts Hamidah has reposted recently: Thais Castello Branco.

Why it matters here

Her agenda shows where the best-funded competitor wants to go next: edit sequences, design trajectories and version history, which is the same differentiated asset Contra Labs captures as screen-recorded trajectories. Her background in data economics and consent is directly relevant to how Julian structures licensing with creatives. The Prototype fellowship is also an open door: it offers Taste Labs datasets and grants to outside researchers.

How to reach

Publicly reachable through the Prototype fellowship application and the research contact on tastelabs.com; also on X (@didaoh) and GitHub. An academic or technical question works better than a sales pitch, for example which of the research problems they consider hardest, or a trajectory-capture methodology exchange. Speculative: she may respond to data-economics or consent framing given her publication record.

What we could not establish
  • Exact start date at Taste Labs
  • Whether she has graduated from McGill
  • Her specific ownership (research lead vs. individual contributor)
  • The X handle evidence comes from her 2024 IFP author page; current activity on it was not checked

Sources

  1. Taste Labs: Requests for research (Hamidah Oderinwale, 16 Aug 2026)
  2. Taste Labs: Introducing Prototype (Hamidah Oderinwale, 16 Aug 2026)
  3. IFP author page
  4. Topos Institute profile
  5. GitHub profile
  6. Semantic Scholar author search (publication list)
Also at Taste Labs
high confidence

Thais Castello Branco

Founder & CEO