Unified linguistic data sourcing, annotation, and model evaluation — 22 Indian languages, freelancer-scale, on-prem.