Your First Workbench Project

Create a project, notebook, and run your first AI-assisted tool call.

The MDM Workbench lets you manage data quality workflows through an AI assistant that can call Golden Suite tools on your behalf. This guide walks you through creating your first project, starting a notebook, and running a tool call.

Prerequisites

A Golden Suite account with Workbench access
At least one data source uploaded (see the Entity Resolution Workflow guide for ingestion steps)
A modern browser (Chrome, Firefox, Safari, or Edge)

1. Create a Project

Navigate to /golden/projects and click New Project. Fill in the following fields:

Name — A short, descriptive name for this project (e.g., "Customer Dedup Q2")
Description — What you're trying to accomplish (e.g., "Deduplicate and merge customer records from CRM and billing exports")
Trust Level — Start with Standard. This controls which tools the assistant can run automatically vs. which require your approval.

Click Create to set up the project.

2. Create a Notebook

From the project detail page, click New Notebook. Notebooks are the primary workspace — each one is an event-sourced log of every action taken, so you have a full audit trail.

Give the notebook a name (e.g., "Initial profiling") and click Create. You'll land in the workbench view with three panels:

Left — Issues and data preview
Right — Chat panel for interacting with the assistant
Bottom — Event log showing every action taken in this notebook

3. Chat with the Assistant

In the chat panel on the right, type your first message:

"Profile the sample customer data and tell me about any quality issues"

The assistant will determine which tool to run (in this case, run-goldencheck) and present the tool call for your review. Depending on your project's trust level, it may run automatically or ask for your approval first.

4. Approve and Review

If the trust level requires confirmation, you'll see the proposed tool call with its parameters. Click Approve to execute it.

Once the tool runs:

Results appear directly in the chat as a formatted summary
The event is recorded in the event log at the bottom of the screen
Any issues found (nulls, duplicates, format violations) appear in the left panel

Review the profiling results. The assistant will highlight the most significant quality issues and suggest next steps.

5. Continue the Workflow

Build on the profiling results by sending follow-up messages. For example:

"Map these columns to the canonical customer schema"

"Clean the data using the mapping you just created"

"Run dedup with a 0.90 threshold on name, email, and address"

Each tool call adds to the notebook's event log, creating a full history of your data management workflow.

Tip: Check the event log regularly. It shows every tool call, its parameters, and results. If something looks wrong, you can trace back through the log to find exactly where the issue was introduced. The event-sourced design means nothing is ever lost.

← Entity Resolution Workflow Source Connectors →