A dataset you can inspect.
A practical guide to defining, reviewing, and receiving a custom dataset. Final deliverables and service levels are set in your project agreement.
Start with a collection specification
Describe the target task, modalities, operating conditions, and coverage gaps. Specify languages, speaker or scene diversity where appropriate, exclusion criteria, and anticipated volume. Agree on an evaluation set before scaling collection.
Choose a delivery format
Common formats include COCO-style JSON for vision, WAV with JSONL annotations for audio, timestamped manifests for multimodal samples, and JSONL or Parquet for text and preference data. Field names, storage layout, and partitioning are agreed during scoping.
{
"dataset_id": "project_example",
"version": "1.0.0",
"split": "train",
"sample_id": "sample_000142",
"source_ref": "source_007",
"rights_ref": "agreement_001",
"annotation_ref": "annotations/sample_000142.json",
"review_state": "accepted"
}This schema is illustrative. It is not a public API contract or a sample from an available dataset.
Define acceptance before collection
- Coverage: agreed distributions, environments, and minimum sample counts.
- Annotation: ontology version, reviewer calibration, and disagreement handling.
- Validation: missing fields, duplicates, corrupt files, and modality alignment checks.
- Evaluation: agreed sampling method, quality rubric, and remediation process.
- Delivery: format, version identifier, dataset splits, and handoff documentation.
Keep sourcing and rights explicit
Define the permitted sources, contributor consent requirements, personal data handling, and permitted uses before collection starts. Source references and licensing records should accompany the dataset as agreed in the project specification. A dataset preview on this website does not grant a training-data license.
Price the actual scope
Custom projects are scoped around collection difficulty, contributor requirements, annotation complexity, review effort, volume, and delivery timelines. Submit a project brief to establish those requirements. This website does not list fixed prices or guaranteed turnaround times.