Ask the AI Assistant your first question (5 min)

By the end of this Tutorial

• You'll know where in the analyst flow the AI Assistant is most effective.
• You'll run one cycle of turning a single natural-language question into three outputs — data location, chart, and code.

Prerequisites

You need portal access plus Reader access to at least one collection and dataset for the results to be meaningful. If you don't have a collection of your own, finish Create your first collection (5 min) first.

1. Open the Assistant (30 seconds)

• Click the AI Assistant button in the upper-right, or use the quick shortcut Cmd/Ctrl + K.
• When the chat panel opens on the right, the input box is ready for your first question.

2. Question 1 — Finding data (1 min)

"Are there any revenue datasets from last quarter?"

The Assistant answers in one of two ways:

• If a matching dataset exists within your permission scope, it suggests links as cards. Click a card and you jump to the dataset detail.
• If none exist, it recommends similar candidates (matching on tags and descriptions) or suggests refining the search keywords.

Worth remembering: the Assistant only points to resources you have access to. Resources you can't see are excluded from recommendations too.

3. Question 2 — Building a chart (2 min)

With the dataset you just found (or one in your own collection) selected:

"Make a bar chart of revenue by category from this dataset."

The Assistant surfaces a widget-creation action in the Suggested actions panel. Don't execute it right away — the recommended order is:

1. Confirm that the recommended X-axis and value field mapping matches your intent.
2. Click Apply changes. If the rendered chart looks empty, check the NULL ratio in the input data.
3. If the chart type isn't a fit, re-request a different type from the same suggestion box.

The Assistant's answer is always a hypothesis to verify. A line chart may fit better than a bar chart, and average may be a better aggregate than sum.

4. Question 3 — Generate code (1 min 30 sec)

When you need a transform or analysis snippet, open the Generate code panel with ⌘I (or Ctrl + I).

"Give me a SQL query that totals revenue by category in this dataset."

Read the generated code once before Insert, and use Retry to refine it. The output comes in a form you can drop straight into a code node (Engineer Path).

Next steps

• To make the Assistant's answers more accurate, register internal PDFs, Notion pages, and reports as Knowledge to fill the RAG context. The related user docs page is Knowledge / AI Chat.
• The place the Assistant sits in the analyst's overall flow is summarized one more time in lesson 06 of the Analyst Path.