Introduction to Synthetic Data Generation
Generate synthetic goldens from your own data sources
Overview
Synthetic data generation allows you to automatically create high-quality goldens from your existing data sources — documents stored in Google Drive, messages in Slack channels, pages in Notion, or files in SharePoint. Instead of manually writing goldens one by one, you can connect a data source and let Confident AI generate evaluation-ready goldens at scale.
How It Works
At a high level, synthetic data generation follows three steps:
Connect a Data Source
Navigate to Project Settings > Data Sources and connect an external source such as Google Drive, Slack, Notion, or SharePoint. Each source type requires its own set of credentials.
Create a Generation Config
Under Datasets > Automations, create a generation configuration that points to your data source. You can control parameters like the maximum number of goldens generated per context chunk.
Generate
Click Generate and Confident AI will pull documents from your data source, chunk them into contexts, and use an LLM to synthesize goldens — complete with inputs, expected outputs, and context fields.
Supported Data Sources
Google Drive
Connect a shared folder and generate goldens from .txt, .pdf, and .docx files.
Slack
Generate goldens from channel message histories.
Notion
Pull page content and generate goldens from your knowledge base.
SharePoint
Connect via Azure AD and generate goldens from SharePoint files.
Next Steps
Data Source Connectors
Learn how to set up credentials and connect each supported data source.
Manage Datasets
Learn how to review, finalize, and manage the goldens that were generated.
Last updated on