Datasets

Annotate regions, transcribe with a vision model, export JSONL for fine-tuning.