Describe with AI: Tables & Dashboards

Overview

The AI-Powered Auto-Description feature in CastorDoc revolutionizes how contributors document dashboards and tables. Previously, AI-generated suggestions for asset documentation were only available if a SQL source query was present. With this new feature, the AI leverages all available metadata—not just SQL queries—to generate rich, accurate, and context-aware descriptions for dashboards and tables.

Why This Matters

• Broader Coverage: Previously, only assets with a SQL source query could benefit from AI suggestions. Now, assets without SQL queries—such as most dashboards and many tables—can also be auto-documented using their metadata.

• Efficiency: Contributors can document assets quickly and completely, without needing to write content manually or rely on external tools.

• Scalability: This unlocks auto-descriptions for up to 40x more dashboards (from 8,000 to over 309,000), plus similar gains for tiles and viz models.

How It Works

1. Metadata-driven AI Suggestions

When you ask the AI to suggest a description for a dashboard or table:

• The AI gathers all relevant metadata associated with the asset, including:

  • Dashboard/Table name, path, and type

  • Technology (e.g., PowerBI, Looker)

  • Tags and labels

  • Editor and owner information

  • IsVerified / IsDeprecated status

  • Folder path and source system

  • Descriptions (internal and external)

  • All source SQL queries (if any)

  • Parent and child asset relationships

  • Column/field names, types, and descriptions

  • Frequently used users, pinned assets, mentioned assets

  • Additional context such as report page names, business objects, and data update frequency

2. No SQL? No Problem!

• If an asset has no SQL source query, the AI uses all other available metadata to generate a high-quality description.

• For dashboards, the AI leverages the same context used in the Dashboard Q&A feature (including information from tiles, not just the dashboard itself).

• For tables, all SQL queries associated with the table are used (not just the primary one), along with rich column and relationship metadata.

3. User Experience

• AI Suggestion Button: When documenting an asset, click the AI button to generate a suggested description.

• Multiple Queries: If multiple queries are available, the AI analyzes them all automatically—no need for a modal asking you to choose.

• Preview & Edit: Review, accept, or edit the suggested description before saving.

Example Use Case

Suppose you have a PowerBI dashboard called "GHG Monitoring Dashboard - PROD" with no SQL source query. When you click the AI button, the system will:

• Use the dashboard’s metadata: name, description, folder path, tags (e.g., "domain:finance"), verification status, source system ("PowerBI Corp"), and more.

• Include information about editors, frequent users, owners, and related business objects or pinned assets.

• Reference parent tables and fields, metrics definitions, update frequencies, and other context.

• The AI will then generate a comprehensive, readable description, summarizing the dashboard’s purpose, content, data sources, and key metrics.

Supported Metadata (Not Exhaustive)

• Dashboards: Name, folder path, descriptions, tags, technology, type, source, editors, frequent users, pinned assets, owners, parent/child assets, page names, etc.

• Tables: Name, database/schema, technology, columns (names, types, user/external descriptions), joins, all source SQL queries, parent/source tables, verification/deprecation status, editors, owners, tags, pinned/mentioned assets, etc.