We are hiring people who have been wrong in public.

You do not need a background in machine learning. You need to have spent a long time on a search that did not work and to know precisely why it did not.

What we read for

One result, in detail, beats a list. Tell us about a single thing you tried to establish — what you expected, what the check was, and what came back. A paragraph is enough. We are trying to see how you decide something is true, not how much you have done.

Negative results are welcome and specific ones are better. "It didn't generalise" tells us nothing. "It didn't generalise, and it turned out our held-out set shared a preprocessing step with training" tells us a great deal.

No take-home puzzles. The process is a conversation about your result, a conversation about one of ours, and a working session on a real search space. We will tell you what we cannot answer yet, because that list is public anyway.

Open roles

Six seats, based in Hong Kong, remote-friendly. Titles are a starting point — if you fit two of these better than one, say so.

  • Research Scientist — Hypothesis generation

    Ranking and calibration. Hong Kong · remote-friendly

    Apply
  • Research Scientist — Verification and evaluation

    Building the half of the loop that can say no. Hong Kong · remote-friendly

    Apply
  • Systems Engineer — Search infrastructure

    Run orchestration, logging, reproducibility. Hong Kong · remote-friendly

    Apply
  • Research Engineer — Simulation and tooling

    Simulators, proof checkers, prior-art retrieval. Hong Kong · remote-friendly

    Apply
  • Experimentalist — Instruments and wet lab

    You have run the machine yourself. Hong Kong

    Apply
  • Research Intern

    One thread, one term, one honest write-up. Hong Kong · remote-friendly

    Apply

Nothing here fits?

Write anyway. If you can describe a search space we should be running against and why it would be hard, that is a better application than any of the titles above.