Create a fine-tuning dataset proposal
Start a CrewAI run that assembles domain knowledge into Q/A examples.
Crew events are published under the job’s UUID; the portal polls
/api/events/{job_id}. The response carries that ID as X-Correlation-ID.
Authorizations
The access token received from the authorization server in the OAuth 2.0 flow.
Body
Start a dataset-generation run for a DataHub domain.
DataHub domain name or URN whose knowledge seeds the dataset.
512Base model this dataset will fine-tune. Must be present in FINETUNE_BASE_MODEL_ALLOWLIST; validated server-side.
256Human-friendly dataset name. Defaults to '-finetune'.
256Free-form guidance and facts the user wants baked in — extra context the crew should weave into the training examples.
20000Soft target for how many Q/A examples to generate. Clamped by FINETUNE_DATASET_CREW_MAX_EXAMPLES.
1 <= x <= 500Owner username. Ignored for non-admin callers (the authenticated user is always the owner); admins may set it to create on behalf of another user.
256Response
Crew run started. Poll the job for the proposal.
The crew's proposed dataset, returned for review.
The full example set lives in S3 (proposed_dataset_s3 / after approval,
approved_dataset_s3). sample_examples is a bounded preview.
Lifecycle of a fine-tuning dataset job.
running, pending_approval, failed, approved, materializing, complete How much of the domain the crew drew on.

