These datasets are built for fairness benchmarking, bias testing, and model robustness evaluation — all aggregated, no PII, no scraping. Each panel combines regional demographic data (age, gender, income, education, occupation) with geographic context to help labs test model outputs across diverse populations.

We’ve just opened a free 30-day pilot for startups, researchers, and AI teams who want to run evaluations using real, ethical ground-truth data. CSV and Parquet formats are available (US + AU regions to start).

I’d love feedback from anyone working on model evaluation, bias detection, or Responsible AI — especially what additional fields or formats would make this most useful.

https://datalis.app/pilot