Miju Labs

← All buyers · Apple Foundation Models (AFM)

J

Jeffrey Nichols

Research Scientist, AI/ML at Apple (intelligent user interfaces) · Apple Foundation Models (AFM)

medium confidenceTier A · Talk to firstSecond-wave labsUI

A career researcher in automatic UI generation — his CMU thesis was literally about automatically generating high-quality user interfaces — now a research scientist in Apple's AI/ML group and a co-author on both of Apple's CHI 2026 UI papers. He also co-created RICO, the mobile UI dataset that much of the field's UI-understanding work was built on, which makes him one of the few people in this index with a track record of shipping the datasets others train on.

Background

PhD in human-computer interaction at Carnegie Mellon under Brad Myers, on automatically generating high-quality user interfaces for appliances, after a BS in computer engineering at the University of Washington. He spent about seven and a half years as a research staff member at IBM Research–Almaden working on web usage, end-user programming, crowdsourcing and social media — crowdsourcing experience that is directly relevant to how human-data collection is designed — then was a staff research scientist at Google on Fuchsia, before joining Apple.

With Ranjitha Kumar's group at UIUC he co-created RICO, the large mobile UI dataset. At Apple he sits in the AI/ML organisation working on intelligent user interfaces, and appears on the UICoder line of work (fine-tuning LLMs to generate UI code via automated feedback) as well as the CHI 2026 designer-feedback and UI-component papers. He was also a CHI 2026 reviewer.

What they run now

UI generation and UI understanding inside Apple's AI/ML group. Two strands matter: UICoder, which improved UI code generation using automated feedback rather than humans, and the CHI 2026 designer-feedback work, which showed that designer critique — comments, sketches, direct manipulation, with rationale — outperforms ranking-based preference data for teaching models to produce well-designed UIs. The pairing is the interesting part: the same group has tried both the cheap automated route and the expensive human route, and published that the human route wins on design quality.

What they buy

Inference. Study-scale designer panels and UI annotation, rather than industrial data volumes. His history with RICO and with IBM crowdsourcing research means he has personally built and run data-collection pipelines, so he will evaluate a vendor on protocol design and annotation quality, not on headcount. Apple names no external human-data vendors publicly.

Career

  1. presentResearch Scientist, AI/ML (intelligent user interfaces), Apple
  2. —Staff Research Scientist (Fuchsia), Google
  3. —Research Staff Member, IBM Research–Almaden — ~7.5 years; crowdsourcing, end-user programming, web usage.
  4. —Carnegie Mellon University, Ph.D., Human-Computer Interaction (advisor Brad A. Myers)
  5. —University of Washington, B.S., Computer Engineering

On the record

Automated vs human feedback · 2026-02

His group has published both automated-feedback (UICoder) and designer-feedback approaches; designer critique won on design quality. source

Interviews & talks

Why it matters here

Useful, with a caveat. He is a genuine believer in UI datasets as research infrastructure and a co-author of the paper that validates Julian's thesis — and unlike Swearngin he is still at Apple. But he is a research scientist inside the most closed lab in the index, and a vendor relationship would have to route through Apple procurement and internal design staffing, which no public source illuminates.

How to reach

Speculation: treat him the same way as Barik — engage on the dataset, not the contract. He co-created RICO, so a credible opener is what a RICO-era mobile UI dataset would look like today if it carried designer critique and rationale instead of just screens and view hierarchies, plus an offer to share a pilot slice. Conference contact (CHI) and his personal site are the channels; he keeps no public social presence I could verify.

What we could not establish
  • Dates for the IBM, Google and Apple roles.
  • Whether he has any influence over Apple's external human-evaluation vendors, or only over research studies.
  • No verified X or LinkedIn profile; the Scholar ID came from a search result rather than his own site.

Why they are on the map

Co-author on designer-feedback UI-generation paper (CHI 2026) source — evidence dated 2026-02

Shapes what 'good data' means.

How Apple Foundation Models (AFM) buys data

AFM 2025 report (Jul 2025): collected large text-image preference data to train a reward model; human graders disagree on 20-30% of pairs; 'internal and external' human eval, vendors unnamed. AFM v3 (Jun 2026) uses in-house human graders. Apple researchers (CHI 2026) collected 1,500 designer annotations from 21 internal designers to fine-tune UI-generation models - direct design-feedback data work.

Dec 2025: Giannandrea retired; Amar Subramanya (VP AI) under Craig Federighi runs AFM, ML research and safety; Zhifeng Chen leads the foundation-models team. Human grading appears in-house; HCI/UI research group in Seattle (Nichols, Barik, Swearngin).

Sources

  1. Personal site — Jeffrey Nichols (Apple AI/ML, career, RICO, CMU PhD)
  2. Apple ML Research — Designer feedback paper page
  3. Apple ML Research — UICoder
  4. Apple ML Research — Apple at CHI 2026