Generic search checklist

Set up a Google Cloud project, turn on Vertex AI Agent Builder, and set up access control for your project. You can use an existing Google Cloud project if you have one already.

Actions

1. Review Before you begin and confirm that you have completed the steps.

Determine what kind of data you'll make available in your search app and prepare it for importing to Vertex AI Search.

You can base your data store on the following types of data:

• Unstructured data. Documents (such as PDFs, HTML files, and TXT files) that are stored in Cloud Storage. Optionally, you can provide metadata in a JSON file or in a BigQuery table.
• Structured data. Data provided with a specific schema. For example, you can provide data in a BigQuery table, as JSON files in Cloud Storage, or from third-party connectors such as Jira.

Actions

1. Review the information about supported data and the relationship between apps and data stores in About apps and data stores.

2. Prepare your data according to the requirements in Prepare data for ingestion.

3. If you need to set up access control to limit the data that users can view in your search app's results, review the prerequisites and follow the instructions for your identity provider type and data source in Use data source access control.

Create a data store and then import your data into it, or set up a third-party connector as your synced data source.

How you import your data depends on where you're importing it from. For example, if your data is in Cloud Storage, you can import it using the console or the API by providing the bucket location of your data.

Actions

1. Follow the instructions for your data source in Create a search data store.

Create your generic search app and connect it to your new data store.

Actions

1. Create a search app.

Vertex AI Search provides many configuration options. Some options depend on whether you plan to deploy a search widget.

Actions

Depending on your use case and whether you plan to deploy the out-of-the-box search widget or integrate search API calls into your own code, Vertex AI Search provides several options for configuration.

1. You can embed a search widget into your website. The widget automatically provides a search bar and expandable search interface. If you plan to deploy the search widget, configure the following:

   • Search widget results. See Configure results for the search widget.

   • Search widget facets (Preview). See Configure facets for the search widget.

2. If you plan to integrate search API calls into your server or application instead of using the widget, you can configure your search settings using the following options:

   • Field settings. For structured data or for unstructured data with metadata, update field settings to refine how Vertex AI Search uses metadata for search. See Configure fields for search.

   • Autocomplete. Depending on your data, set up autocomplete suggestions based on document content, fields, search history, or user events. See Configure autocomplete.

   • Serving controls. Control when search results are boosted, buried, filtered, or redirected, or whether certain queries are associated with other queries. See Configure serving controls.

   • Search tuning (Preview). Tune the search model with your own training data. See Improve search results with search tuning.

   • Custom embeddings (Preview). If you've created your own embeddings, you might prefer to use them instead of those generated by Vertex AI Search to enrich your searches with additional context. This feature is available for data stores with structured data or unstructured data with metadata. See Use custom embeddings.

If you plan to deploy your app by integrating search API calls into your own code, Vertex AI Search provides additional options for configuring how your search results are returned.

Actions

1. Configure your search results with the following options:

   • Filter generic search for structured or unstructured data.
   • Get snippets, extractive answers, or extractive segments. See Get snippets and extracted content.

You can preview your search results to check if your app configurations are working as expected.

Actions

1. To preview your search results, use the console or the API.

   • Console. Use the Vertex AI Agent Builder console Preview page to preview how search widget configurations affect your results. See the Console instructions in Get search results.

   • API. If you're integrating API calls into your application, make API calls to preview your search configurations. See the REST instructions in Get search results.

When you're happy with the preview version of your search app, share it with your users by deploying it to your website.

Actions

1. You can deploy your search app in either of the following ways:

   • Embed the search widget into your website. Vertex AI Search provides code that you can copy into your website or web application. This deploys the search widget. You can preview search results in the console. See Add the search widget to a webpage.

   • Integrate search API calls into your server or application. For full control over how your search results are displayed, you can integrate API calls into your server or applications. For more information about making API calls, see Get search results. For client library resources, see Vertex AI Agent Builder client libraries.

You can maintain your app to ensure that latest and necessary data is available in your data store.

Actions

1. Refresh your data.