ad
ad
Topview AI logo

Create a Search App for BigQuery Dataset in 6 mins #googlegemini #genai #chatgpt #ragai

People & Blogs

Introduction

In today's tutorial, we'll be creating a search engine that leverages a BigQuery dataset using Vertex AI Agent Builder. We will specifically focus on utilizing the Google Cloud release notes dataset to illustrate the steps involved.

Step 1: Defining Our Data

First and foremost, it's essential to determine the dataset we want our search engine to operate on. For this exercise, we have imported the public dataset known as the Google Cloud release notes. This dataset contains vital release information regarding all Google Cloud services, including:

  • Description of release: Contains textual details about what has been released.
  • Release type: Identifies whether the release is a fix or a new feature.
  • Publication date: When the release occurred.
  • Product ID and name: Details of the product (e.g., BigQuery) associated with the release.
  • Version name: Indicates the specific version that was released.

To build our search engine efficiently, we will mainly focus on the product name, publication date, and the description fields.

Step 2: Copying the Table ID

Next, we will navigate to the dataset details within BigQuery and copy the table ID. This table ID will help us seamlessly connect our data source to the Vertex AI Agent Builder.

Step 3: Setting Up Vertex AI Agent Builder

  1. Open Vertex AI Agent Builder and navigate to the Data Stores section.

  2. Create a new data store with the following details:

    • Native source: BigQuery.
    • Data store type: Select structured BigQuery table with schema synchronization frequency set to "one time".
    • BigQuery path: Paste the table ID that you copied earlier.

  3. Click on continue to review the schema. Here, ensure that the description field is set as retrievable, as this will contain the search content. Leave the remaining fields unchanged and continue.

  4. Give your datastore a name, such as "Release Notes Data", and create it.

The data store will start indexing, so allow a few minutes for this process to complete before we proceed to create the search engine application.

Step 4: Creating the Search Engine App

  1. Navigate to the app section and create a new app.
  2. Select the application type as "Search content" and choose the "Generic" option.
  3. Opt for the Enterprise Edition features and advanced LLM features, where you will name the search engine "GCP Release Notes" and set your company name (for example: "R in LLC").
  4. Your newly created app will now appear, with the option to link it to the data store you have previously created. This creates a connection between your app and the BigQuery dataset.

Step 5: Viewing and Configuring Your Data

Once the app is created, navigate to the data section to view the imported data. Check that all the desired columns — specifically the description and product name — are present.

  • Next, head to the configurations section to configure your search preferences:
    • Search Type: Choose "Search and Answer".
    • Generative Summary: Enable the summary feature above search results, set to display five top results.
    • Model Selection: Use "Gemini 1.5 Flash 1 V2" for optimal speed and quality.

Choose to automatically detect the language and disable suggestions for related questions, as the data is strictly pertaining to release notes. Enable feedback options and save your settings.

Finally, go to the preview section and test your search capabilities. For example, input the query “Does Composer support private service connect endpoints?” In response, you should see relevant answers grounded to the release notes, confirming the effectiveness of your search engine.

With these steps, you have successfully built a search engine for Google Cloud release notes using Vertex AI Agent Builder.

Keywords

  • BigQuery
  • Vertex AI
  • Search Engine
  • Google Cloud Release Notes
  • Generative AI
  • Data Store
  • Dataset
  • Configuration
  • Gemini 1.5

FAQ

Q1: What is Vertex AI Agent Builder?
A1: Vertex AI Agent Builder is a tool that allows users to build AI-driven applications, including search engines, that connect to various data sources like BigQuery.

Q2: How do I create a data store in Vertex AI?
A2: You can create a data store within Vertex AI by selecting a native source (e.g., BigQuery) and providing the necessary table ID and schema settings.

Q3: What is the significance of the "retrievable" setting in data configurations?
A3: The "retrievable" setting ensures that specific fields, such as descriptions, can be fetched and displayed in search results, which is crucial for effective search functionality.

Q4: Why should I use the Gemini 1.5 model?
A4: The Gemini 1.5 model is preferred for faster processing and quality responses when generating summaries or search results, making it suitable for applications needing quick results.

Q5: Can I disable related question suggestions in my search app?
A5: Yes, you can disable related question suggestions if your dataset is focused and should not include potentially inaccurate follow-up questions.