Remedi

• 

  live datahub queue

• 

  proposal

• 

  generated artifact

Data quality alerts tell teams that something broke. They rarely provide the context needed to repair it safely.

Remedi turns one failing DataHub assertion into a reviewable repair workflow:

1. Read the affected entity, schema, owners, query, and lineage from DataHub.
2. Generate grounded dbt, SQL, orchestration, or ML repair artifacts from that context.
3. Seal the plan and artifacts with SHA-256 hashes.
4. Require explicit human approval before execution.
5. Write a pending-validation status back to DataHub and wait for the real pipeline rerun.

DataHub is both the context source and the system of record. Remedi never marks an assertion as passed simply because a patch was generated.

The current showcase was verified against a local DataHub OSS instance with a disclosed synthetic failing assertion. That run traversed 35 downstream and 69 upstream lineage nodes, produced seven review artifacts, and made 111 DataHub tool calls. The repository also includes 47 tests and a 27-check self-test covering provenance, approval, write-back, and failure handling.

Live mode fails closed when DataHub or a required provider is unavailable. Fixture mode is isolated and clearly labeled; local integration evidence is not presented as production deployment evidence.

Built With

• datahub-agent-context-kit
• datahub-oss
• datahub-skills
• docker
• fastapi
• openai-api
• playwright
• pydantic
• python
• typer