Concept | Adding a governance layer to Dataiku items#

The Govern node syncs Dataiku items from connected nodes. However, this act of centralization is only what the Govern node does out of the box.

The primary user action is to actually “govern” these synced Dataiku items. When you govern an item, you initialize a way to manage, oversee, and monitor it. Let’s investigate the process for adding a governance layer to an item in the Govern node.

Initiating a governance layer#

You can initiate a governance layer for a Dataiku item from the Governable items or registry pages. Look for the gavel (Gavel icon.) in the row of the item that you want to govern.

Important

Alternatively, you might begin the lifecycle of a project in the Govern node itself — instead of adding it to an existing Dataiku item. In this case, you can create a Govern project first using the Create button on the Governed projects (Governed projects icon.) page. Then later add a Dataiku project to it. You’ll see this in Tutorial | Governance lifecycle.

Once you initiate the governance layer, you have several options for defining how to govern the item. Below is a screenshot of one possible governance configuration. The sections below analyze the various options shown here with reference to these markers.

Dataiku screenshot of the dialog to govern a Dataiku project.

Choosing a governance template#

When adding a governance layer to a Dataiku item, you apply a certain governance template. This may be the standard template available to all users. If your organization holds advanced licenses, you may also have custom templates from which to choose.

In the example above, the standard template will govern the project (see marker 1), but various custom templates will govern the child items (see marker 2).

See also

You’ll see the role of these templates in Concept | Governed items.

Choosing the level at which to govern#

In this example, the user will govern the project and all child items. What exactly are child items?

Dataiku users will be familiar with the information hierarchy inherent to Dataiku items. Starting with the most granular items, it should come as no surprise that:

  • A bundle belongs to a project.

  • Saved model versions belong to a saved model, which belong to a project.

  • Should you have them, the same principles apply to GenAI elements:

    • Fine-tuned LLM versions belong to a fine-tuned LLM, which belong to a project.

    • Agent versions belong to an agent, which belong to a project.

    • Augmented LLMs versions belong to an augmented LLM, which belong to a project.

In all cases, the project is the top-level Dataiku item in the hierarchy. When thinking about your governance strategy, you have a choice to make about how deep into this information hierarchy you’d like to govern.

  • If you govern at the project level, you have the option of extending this layer to include child items.

  • If you start from the child items, such as bundles or model versions, you’ll effectively be forced to govern the parent items.

In other words, your organization may decide to govern only projects. However, it isn’t possible to govern only child items like saved model versions without also governing the parent model and project.

In the example above, this user will govern at all possible levels. Note the three checked boxes for synced child items (see marker 3), but the three unchecked boxes for existing child items (see marker 4).

Tip

If you’re just getting started, try governing at the project level, including all child items, to understand what value it might bring. Later, you can restrict your governance activities to the project level, should you so choose. Ultimately, organizations will need to decide what works best for their own needs!

Governing existing vs. future items#

Markers 3 and 4 introduce the question of existing vs. future synced items.

In addition to thinking about the depth of your governance efforts, you also need to consider the temporal component to governance. The dialog for governing a project enables you to make different template selections for existing and synced (future) child items.

In this example, it may be that any existing child items in the project (marker 4) were simple tests, and therefore don’t need governance. This user only wants to govern items moving forward (marker 3).

Instance rules set by an administrator simplify this process. For example, your organization may define a rule to always govern future child items in any project. Then, a user can override these rules if needed.

Organizing under business initiatives#

Finally, you have the option to organize your governance activity under a business initiative (see marker 5). Business initiatives serve as an optional extra layer of organization to group governance items under a shared theme.

Next steps#

Now that you’ve seen how to add a governance layer to Dataiku items, learn about the resulting objects in Concept | Governed items!