Skip to main content
Context Extraction is a feature within WisdomAI’s Adaptive Context Engine (wACE) that helps you build your Domain’s context by using existing documentation. By uploading files like dbt YAML, internal docs, PDFs, or query logs, the Context Extraction tool proposes reviewable suggestions tailored to your data. You stay in control by accepting, rejecting, or editing suggestions before anything is published. This guide will walk you through the steps to upload your files, extract metadata, and review the resulting suggestions.

How it works

The process follows three stages to ensure all metadata is verified before it is added to your Domain.

Upload

Upload files directly from your Domain page. You can optionally provide a short prompt to steer what the system should focus on during the context extraction.

Extract

WisdomAI analyzes the content and proposes structured items like descriptions, metrics, joins, and business context. When possible, WisdomAI validates these suggestions, for example, by utilizing a test SQL execution.

Review and Publish

Review all proposed changes in the Suggestions experience. Compare the suggestion side-by-side with the source file to ensure accuracy. You can choose to Accept, Reject, or Edit any item before publishing it to your knowledge base.
This feature reduces manual entry and helps standardize definitions across teams while keeping you in control through a review-first workflow. It also improves trust with source attribution by showing exactly which part of your documentation generated a suggestion.

Extracted metadata types

Context Extraction identifies the following categories of information:
Item TypeDescription
DescriptionsTable and column definitions.
MetricsMetric definitions and associated SQL.
JoinsRelationships and join logic between tables.
Business ContextRules, definitions, conventions, and SQL hints.
Reviewed QueriesExample question-to-SQL pairs for model training.

Supported file types

You can use various files to provide context, including:
  • YAML Files: Such as dbt-style configurations.
  • LookML: Looker modeling files.
  • Unstructured Documents: PDFs and text files.
  • Query Logs: Historical SQL records.

Target users

The context extractor feature is available to:
  • Data Admins who manage and curate trusted domain knowledge (descriptions, metrics, joins, and definitions).
  • Explorers who rely on accurate context to ask better questions and interpret results with confidence.

FAQs

No. Extracted items remain as suggestions until an admin clicks Accept in the Suggestions experience.
Suggestions are located in the Knowledge tab of your Domain. Each entry shows the specific source content used for the extraction.
Yes. During the upload step, you can use the prompt field to tell Wisdom exactly what metadata to look for.

Next steps