Stage
Read the uploaded file or bucket path and record what was read.
ML · ships today
A dataset you
can train on
Clean, deduplicate, decontaminate, redact and register a version training can pin.
Ships as dataset_readiness/import
Import a file or bucket, clean it, remove anything that overlaps your evaluation sets, redact personal data, and register a hash-pinned version.
The pipeline
Clean
Normalize rows, remove duplicates and redact PII.
Decontaminate
Remove overlap with your evaluation set so scores stay honest.
Register
Publish a hash-pinned version under the name you choose, with its lineage.
What the record keeps
The source reference, the quality findings, the redaction report, and the content hash and lineage of the version you published.
Honest evaluation
Evaluation overlap is removed before training ever sees the data.
PII handled first
Personal data is redacted before the dataset is registered.
Pin it and move on
Training refers to an exact version, not a path that can change.
Ships as dataset_readiness/import. Checks cover text and tabular data; connectors depend on the installed operators.
See how failure can trigger a reviewed improvement cycle →Run it on your own infrastructure.
Apache 2.0 licensed, self-hosted. cairn pipeline run dataset_readiness/import