Training Data & Curation jobs

Curation became a career the moment everyone admitted data quality moves benchmarks more than architecture tweaks. The job runs from corpus construction — sourcing, deduplication, quality classifiers, contamination checks — to the mixture experiments that decide what a model reads. Synthetic data generation joined the portfolio recently and shows no sign of leaving.

Day to day it's Spark or Ray over enormous text piles, small classifiers trained to grade documents, and ablations to prove a mixture change earned its cost. Licensing literacy is quietly valuable, given where the training data lawsuits went. Titles vary — Data Curation, Data Research, Pretraining Data — so search by the work, not the label.