Challenge 2: Quick No-Code Agentic Applications on Google Cloud using Agentspace

Previous Challenge Next Challenge

 Important This challenge introduces Google Agentspace, a powerful platform for quickly building conversational AI applications without deep coding expertise. Focus on integrating your prepared data and enabling natural language interaction.

Introduction

With your data now centralized in BigQuery, the next step is to make it accessible and intelligent. Google Agentspace allows you to quickly deploy conversational AI applications that can understand natural language queries and retrieve information from your various data sources. This means users can simply ask questions about your SAP or Salesforce data, and the AI will provide answers, acting as a smart assistant. You’ll also learn to integrate external actions like sending emails.

Description

In this challenge, you will build and configure an Agentspace application to serve as an intelligent assistant for your “ice cream” business. Your tasks include:

  • OAuth Configuration: Set up an OAuth consent screen for internal applications and create an OAuth client for your Agentspace application. This is crucial for enabling secure user authorization, especially for actions like sending emails.
  • Agentspace App Creation: Instantiate a new Agentspace application, selecting the appropriate type and tier (e.g., “Search + Assistant”).
  • Data Store Integration: Connect your Agentspace app to the BigQuery tables and view you created in Challenge 1 (MaterialMasterData, CustomerCases, fda_ice_cream_enforcements). Additionally, integrate an unstructured data source from Cloud Storage (like a recipe PDF).
  • Action Integration: Connect and configure the Gmail tool as an action within your Agentspace app, enabling it to send emails.
  • Google Search Grounding: Enable Google Search grounding to allow your agent to pull information from the web to augment its responses.
  • Interaction and Testing: Once indexing is complete, interact with your Agentspace application using a series of diverse natural language prompts. Observe how the agent leverages your integrated structured and unstructured data, as well as external web search, to provide comprehensive answers and execute actions.

 Note Indexing data stores in Agentspace can take several minutes. Be patient and use this time to explore the interface or review the next challenge. You can also work on Task 3 in a separate tab while waiting.

Success Criteria

  • Verify that an OAuth consent screen and a “Web application” OAuth client are successfully configured, with the correct redirect URI for Agentspace.
  • Demonstrate that your Agentspace application is created and accessible.
  • Show that all required data stores (three BigQuery sources, one Cloud Storage source) are successfully connected and indexed within your Agentspace application.
  • Confirm that the Gmail action is properly configured with the correct OAuth client ID and secret.
  • Successfully interact with the Agentspace app using various prompts that demonstrate its abilities:
    • Prompt 1:
      • Datasource: MaterialMasterData (SAP via BQ)
      • Prompt: Analyze our product portfolio. Identify and summarize the key strengths of our ice cream catalog, focusing on: Flavor variety (range of types offered), Unique or innovative flavor profiles mentioned, Mention of premium ingredients or quality attributes, Presence of interesting textures (swirls, inclusions, crunches). Provide a concise summary of these strengths.
    • Prompt 2:
      • Datasources: CustomerCases (SFDC via BQ), Web Search
      • Prompt: Retrieve customer feedback on our ice cream products, focusing on the improvement suggestions from our customers. Can you identify any particular pattern in this feedback? Additionally, conduct a review our competitors’ strategy online. Based on both results, suggest what should we work on to improve our portfolio and drive sales.
    • Prompt 3:
      • Datasources: CustomerCases (SFDC via BQ), Web Search
      • Prompt: Generate a concept image for a potential new summer seasonal flavor: "Lemon Berry Bliss". The image should show a scoop of pale yellow lemon ice cream generously swirled with a vibrant mixed berry (strawberry, blueberry, raspberry) ribbon. It should look bright, appealing, and refreshing, perhaps set against a simple, clean background.
    • Prompt 4:
      • Datasource: fda_ice_cream_enforcements (BQ View), Web Search
      • Prompt: Analyze historical product recalls related to ice cream and information on the web to find out how we can avoid major issues and recalls with our products and build a great reputation.
    • Prompt 5:
      • Datasources: CustomerCases (SFDC via BQ), MaterialMasterData (SAP via BQ), Recipes PDF (Cloud Storage)
      • Prompt: We have got a request to focus on improvement of the recipe of Pistachio River Ripple. Based on the customer feedback and an existing recipe, suggest what exactly needs to be changed. List exact ingredients we need to purchase to make those changes.
    • Prompt 6: (This uses the Gmail Tool - you may be prompted to authorize)
      • Action: Send Email
      • Prompt: Send an email to our purchasing organization (purchasing@company.com) with the recipe improvement suggestions for Pistachio River Ripple and the ingredients that need to be purchased.

Tips

  • When setting up OAuth, ensure the “Authorized redirect URIs” for your web application client are exact.
  • Give meaningful names to your Agentspace app and data stores for easier management.
  • Experiment with different phrasing for your prompts to understand the agent’s natural language understanding capabilities. Feel free to explore multimodality (e.g. generate images)
  • Observe the “sources” cited by the Agentspace responses to understand which data stores it is utilizing for each query.
  • In the next task, you will need to provide an email with recommended new ingredients. Please, make sure that you request Agentspace to generate this email. Example prompts:

“We have got a request to focus on improvement of the recipe of Pistachio River Ripple. Based on the customer feedback and an existing recipe, suggest what exactly needs to be changed. List exact ingredients we need to purchase to make those changes.”

“Send an email to our purchasing organization (purchasing@company.com) with the recipe improvement suggestions for Pistachio River Ripple and the ingredients that need to be purchased.”

Learning Resources

Previous Challenge Next Challenge