Metadata Reviewer User Manual

Package: ai4data.metadata.reviewer

A multi-agent LLM pipeline for dataset metadata quality assurance.

Detect typos, inconsistencies, invalid content, and missing information across metadata records — automatically and consistently.

This manual is the operational reference for installing, configuring, and running the Metadata Reviewer. For a shorter overview of the pipeline and design rationale, see the Metadata Reviewer overview.

Contents

• Introduction — what the package does and how this manual is organized
• Core Concepts — pipeline, categories, severity, output schema, architecture
• Installation — requirements and provider extras
• Quick Start — build a client, submit metadata, retrieve results
• The Client API — constructor, factories, submission, job management
• Jobs — lifecycle, waiting, cancellation, cleanup
• Advanced Usage — team presets, custom manifests, exclusion rules
• AI Suggestion Review Board — triage pipeline output in the browser
• End-to-End Workflow — catalogue review cycle
• Troubleshooting & FAQ — common problems and fixes
• Quick API Reference
• Glossary
• Default Agents Manifest