AI / ML Engineer
Make the forecast more accurate. Build the layer that decides how much each AI weather model counts, and the scoring that proves it works.
What you'll do
- Design and test ways of combining GraphCast, GenCast, Aurora, AIFS and other models, weighted by variable, region, lead time and weather regime
- Own the accuracy scoring: RMSE, CRPS and skill against ERA5 and the ECMWF IFS baseline
- Improve accuracy first where it matters most commercially — North American severe weather, precipitation and wind
- Publish results honestly, including where we lose, and decide which new models join the ensemble
What we look for
- Strong machine learning background, ideally with weather, climate or other physical-science forecasting
- PyTorch and the scientific Python stack (NumPy, xarray)
- Hands-on experience with at least one AI weather model, or a clear record of learning fast in a new domain
- Comfortable with probabilistic forecasts and uncertainty, not just point predictions
How hiring works
- Step 1
Apply
One page, about five minutes. Your LinkedIn and CV do most of the talking.
- Step 2
Verify your identity
About three minutes: your phone, an ID document and a quick live selfie, run by Didit. Every candidate does this before an interview.
- Step 3
Book a 15-minute call
Once you're verified, we email you a link to pick a time — weekdays, 8am–4pm Eastern Time (ET), at least 24 hours ahead. We read your application before we talk.
- Step 4
Talk and agree terms
The intro call, a technical conversation about work you've actually done, then contract terms, rate and start date in writing.
Pay
$75 – $120 per hour as an independent contractor, depending on experience. We publish the range on every role so you know before you apply, not after three interviews.
About Ensemble Weather AI
We combine many independent AI weather models into one better-calibrated forecast and deliver it as an API — severe storms, nowcasting, air quality and long-range climate. It's early: the ingest pipelines, the ensembling layer and the v1 API are being built now, so the people who join next will own that core rather than maintain someone else's.