LinkedIn
Twitter
With the Looker Dashboard Summarization extension, you can use Vertex AI’s Gemini model to summarize your dashboard data, prescribe next steps, and share AI-powered insights.
Recently, we launched a tile extension feature that enables you to install a Looker extension as a tile within your dashboard. You can already install a Looker extension, a full-fledged application, in your Looker data platform to add functionality like natural language queries, advanced geospatial analytics, and more.
The new tile extension feature enables you to integrate a Looker extension’s functionality with your dashboard’s data and UI, providing you access to Looker extensions without leaving your dashboard or interrupting your workflow. In this post, we walk through the Looker Dashboard Summarization extension’s capabilities, its architecture, and how you can try it out.
What can the Looker Dashboard Summarization extension do?
The Looker Dashboard Summarization extension is an application that leverages Vertex AI to provide 3 core capabilities:
- Summarization: For every query in your dashboard, the extension automatically generates a concise summary of the data, highlighting key trends, outliers, and noteworthy patterns.
- Prescription: The extension provides actionable next steps based on the data, suggesting relevant follow-up analyses or operational actions.
- Action: share your AI-powered insights with colleagues or export them to your preferred business tools, such as Google Chat or Slack, directly from your Looker dashboard.
Let's illustrate these capabilities with a real-world scenario. Imagine you have a Looker dashboard tracking customer loyalty as shown below.
In this example, the Looker Dashboard Summarization extension is installed as a tile in the dashboard on the left side alongside the other dashboard tiles. Next, we’ll examine each part of the extension in the screenshot in more detail.
Summarization
For each of your dashboard’s queries, the extension will describe the query, summarize the data, and highlight outliers in the query’s data. With the extension installed as a tile, it helps you gain insights from your dashboard without needing to drill into each query separately.
The screenshot describes how the “Loyal Customers” query returns a 30-day repeat purchase rate of users and other metadata and then summarizes the query’s data, noting the majority repeat purchase rate and specific outliers.
Prescription
After the extension summarizes each query and the resulting data, it prescribes next steps for each query in your dashboard. The next steps can consist of follow-up operational actions or additional data analysis suggestions and give you an easy starting point to further understand your data.
In our example, the extension suggests investigating outliers in the resultant data and performing further analysis.
Action
Without leaving your dashboard, you can export summaries and next steps from the extension for immediate action. The tile extension enables you to export insights directly to business tools that your organization uses, such as Google Chat or Slack. The extension can also edit summaries and next steps and display a shorter, refined report so you can share the information more easily.
This screenshot shows the three extension buttons that trigger the export and refined report capabilities. What is the extension’s architecture?
The extension makes use of Google Cloud services to enable its dashboard insights. The extension’s architecture consists of:
- A modern React frontend application that runs on the Looker Extension Framework
- A websocket backend service deployed to Cloud Run
- Vertex AI’s multi-modal Gemini model
- Your Looker instance
This architecture diagram describes the flow of data throughout the extension with each numbered step:
- The frontend application sends dashboard and LookML metadata to the backend service to give the Gemini model additional context outside of just the data from the dashboard’s queries.
- The backend service queries the dashboard’s underlying data with Looker’s Query API.
- The backend service prompts Vertex AI’s Gemini model with all the previously mentioned data and uses WebSockets to stream the model’s text response back to the tile extension frontend.
- The frontend application sends the generated text responses to business tools like Google Chat or Slack.
How do I install and run the extension in my dashboard?
Requirements
You will need at a minimum:
- Looker access, either through a Looker original license, an active Looker core trial, or an active Looker core license
- A Google Cloud billing account
- Cloud Run API enabled on the billing account
- Vertex AI APi enabled on the billing account
Steps
For your convenience, you can follow our codelab that covers how to run the extension locally and deploy in production. You can also follow the extension’s repository readme.
At a high level, you will do the following:
- Configure backend service with Looker credentials.
- Deploy the backend service as a Docker image to Google Cloud Run. Optionally, you can deploy the backend service to any other hosting environment that supports websockets.
- Build and deploy the tile extension’s JavaScript frontend as a LookML project.
- Optionally set up the export integrations to Google Chat or Slack.
The codelab and readme also recommend additional steps such as helping fine-tune Gemini’s responses for better dashboard insights.
Contribute to the community 🤝
To get started explore the Looker Dashboard Summarization extension codelab and see how you can leverage Vertex AI’s Gemini model from right within your dashboard.
We encourage you to give us feedback in the repository and contribute changes to the tile extension. We hope this tile extension inspires you to build your own tile extension for your unique dashboard use cases.