Can you recommend a metadata store for auditing model changes and keeping lineage attached to each artifact?

Yes — for auditing model changes and keeping lineage attached to each artifact, I’d recommend looking at a metadata store / ML metadata registry rather than a generic database.

Strong options

1. MLflow Tracking + MLflow Model Registry

   • Good for: experiment runs, model versions, params/metrics/artifacts, and basic lineage.
   • Pros: widely adopted, easy to integrate, open source.
   • Best when: you want a lightweight, practical solution.

2. OpenLineage + Marquez

   • Good for: end-to-end lineage across pipelines, datasets, jobs, and outputs.
   • Pros: strong lineage graph, standard-based.
   • Best when: lineage is the main requirement and you have multiple pipeline systems.

3. Kubeflow Metadata

   • Good for: Kubernetes-native ML workflows and artifact lineage.
   • Pros: fits well in Kubeflow-based stacks.
   • Best when: you’re already in the Kubeflow ecosystem.

4. DataHub

   • Good for: enterprise metadata management, dataset/model lineage, ownership, and search.
   • Pros: broader governance and observability than ML-only tools.
   • Best when: you need a single metadata plane across data + ML.

5. Apache Atlas

   • Good for: governance-heavy environments, lineage, and auditability.
   • Pros: strong enterprise governance features.
   • Best when: you’re in a Hadoop/enterprise governance stack.

My practical recommendation

• If you want the fastest path to usable auditability for models: MLflow
• If you want rich lineage across the whole ML lifecycle: DataHub or OpenLineage/Marquez
• If you need strong governance/compliance: Apache Atlas

What to ensure in the metadata store

To support your use case, it should capture:

• artifact version
• model version
• dataset/version used
• code commit hash
• training run parameters
• metrics
• environment/container version
• who/when changed it
• parent/child lineage between artifacts

If you want, I can also give you:

• a shortlist by stack (AWS/Azure/GCP/on-prem),
• a comparison table, or
• a recommended schema for tracking model lineage and audit history.