Miju Labs

← All buyers · NVIDIA (Nemotron / Cosmos)

J

Jane Polak Scowcroft

Senior Director, Generative AI Data, NVIDIA (title from secondary sources) · NVIDIA (Nemotron / Cosmos)

medium confidenceTier A · Talk to firstSecond-wave labstext

The data executive behind NVIDIA's open models: she appears in the Nemotron 3 report under Leadership, Product, and Evaluation/Safety/Release, and her LinkedIn presents her as building the data that fuels NVIDIA's models. Her whole career is large-scale human data collection — she led Mozilla's Common Voice, one of the biggest crowdsourced speech datasets ever built — which makes her the rare buyer who understands vendor economics and data ops from the inside.

Background

Master's in psychology and a bachelor's in computer engineering — an unusual combination that shows in her work, which is consistently about getting good data out of people at scale. She was head of product at CSIRO's Data61 in Australia, then led Mozilla's Common Voice team, where the open speech corpus passed 13,000 hours; she co-announced an NVIDIA–Mozilla Common Voice release in 2021 as she moved to NVIDIA.

At NVIDIA she started as a product lead for data strategy on conversational AI products, with a stated commitment to equity and diversity through openly available technology. By late 2025 she is listed in Nemotron 3's leadership, product and evaluation/safety groups, and by 2026 third parties describe her as a senior director for generative AI data; a LinkedIn post of hers marks five years at the company.

What they run now

Data strategy for NVIDIA's generative AI models. In Nemotron 3 she sits in three groups at once — Leadership, Product, and Evaluation, Safety and Release — which is the signature of someone who owns data as a product line rather than as a research input: sourcing, licensing and redistribution rights, evaluation data, and what can be published. NVIDIA's stated Nemotron policy is to release weights, software, recipes and all data for which it holds redistribution rights, and someone in her position is the one who negotiates whether a supplier's data can be released that way.

What they buy

Inference, but well-founded: data sourcing and vendor relationships for NVIDIA's generative models, including the licensing and redistribution terms that let NVIDIA publish what it buys. NVIDIA's documented human-data purchases are HelpSteer-style preference annotation through Scale AI and Translated; her remit spans pretraining data, evaluation data and release policy across modalities, which in principle includes Cosmos video and visual work as well as Nemotron text.

Career

  1. 2021 – presentSenior Director, Generative AI Data (earlier: product lead, data strategy for conversational AI), NVIDIA — Five-year milestone noted in her own 2026 LinkedIn post; title from secondary sources.
  2. 2021Lead, Common Voice, Mozilla — Crowdsourced open speech dataset past 13,000 hours.
  3. —Head of Product, CSIRO Data61
  4. —not stated publicly, Master's in psychology
  5. —not stated publicly, Bachelor's in computer engineering

On the record

Open data · 2021-07

Her public record is built on openly available, crowdsourced data — Common Voice and NVIDIA's open dataset releases. source

Interviews & talks

Why it matters here

She is the most plausible economic buyer in the NVIDIA cluster — above the researchers who write the specs, and with a background that means she will not need convincing that human data is a real supply chain. Her Common Voice history also makes her receptive to arguments about contributor treatment, provenance and openness, which is a differentiator a small European vendor can credibly own. The risk is fit: NVIDIA's visible human-data demand is text and speech, not design.

How to reach

Speculation: approach her on data supply and rights, not on taste. The angle most likely to land is that NVIDIA wants to publish what it buys, and most human-data vendors' contracts make that hard — a vendor that can deliver expert visual and design judgement with clean redistribution rights and documented provenance is solving her problem, not just supplying labour. Her Common Voice and Data61 background makes 'openly releasable, well-documented, fairly sourced' the right vocabulary. Verify her current title before any outreach.

What we could not establish
  • Her exact current title — 'Sr. Director, Generative AI Data' comes from third-party LinkedIn posts, and NVIDIA's own author bios are years out of date.
  • Whether her remit covers visual/video data (Cosmos) or only language models.
  • Which vendors she personally manages; none are named in Nemotron 3.
  • Her universities and degree years are not public.

Why they are on the map

Listed under Leadership, Data and Evaluation/Safety in Nemotron 3 report - check if she owns data sourcing source — evidence dated 2025-12

Owns the human-data budget and vendor list.

How NVIDIA (Nemotron / Cosmos) buys data

HelpSteer3-Preference (2025) used 6,400+ annotators in 77 countries sourced/managed by Scale AI (general, STEM, code with degree/SWE bars) and Translated (multilingual). Nemotron 3 report (Dec 2025) lists dedicated Data, Post-training, Evaluation/Safety and Project Management groups but no annotation vendor; RL is mostly environment-based.

Nemotron post-training under Oleksii Kuchaiev (Applied Research), leadership incl. Jonathan Cohen, Bryan Catanzaro, Kari Briski; HelpSteer human-data programme run by Zhilin Wang and co-authors; 15-person project-management pool. Visual generation in Ming-Yu Liu's research group.

Known vendors: scale-ai, Translated

Sources

  1. NVIDIA Nemotron 3 report — she appears under Leadership, Product, and Evaluation/Safety/Release
  2. NVIDIA blog author page — bio: data strategy product lead, ex-Mozilla Common Voice, ex-CSIRO Data61
  3. NVIDIA Technical Blog — Common Voice 13,000 hours release (2021)
  4. LinkedIn — profile headline about building the data fuelling NVIDIA's models