import SocialEmbed from "../../../../components/SocialEmbed.astro";


# How to Automate PR Summaries with Opal AI

<SocialEmbed platform="youtube" id="gphcuJu8iHo" />

## Description

Uncover how Opal from Google Labs revolutionizes building AI systems with its intuitive natural language and visual workflow editor. This tutorial guides you through creating an AI-powered mini-app, Git-Clarity, designed to summarize GitHub pull requests efficiently. Understand Opal’s multi-step workflow creation, fast iteration capabilities, and instant hosting features, making AI development accessible for everyone.

Resources: 
Learn more → https://goo.gle/3Yx4aDq 

Chapters: 
0:00 - Introduction to Opal from Google Labs
1:06 - Opal’s intelligent multi-step workflows
2:10 - Exploring the generated workflow steps
3:29 - Verifying the complete workflow
4:52 - Iteration in practice: Refining output
5:25 - Sharing your live prototype
5:35 - Closing thoughts and future possibilities with Opal

Subscribe to Google for Developers → https://goo.gle/developers 

Speaker: Rody Davis 
Products Mentioned:  Google AI

## Transcript

### Introduction to Opal from Google Labs

**0:02** · Opel is a new experiment from Google Labs that you can use for building and testing chained AI systems in minutes, all using natural language and visual workflows. In this video, we're going to be using Opel to build a fully working mini app called Git Clarity, an AI powered tool that turns any GitHub pull request into a clear one paragraph summary for non-technical stakeholders.

**0:24** · We'll build this entire thing from a blank Opal canvas using only natural language prompts and Opel's visual workflow editor. There's no coding required here. Instead, you describe what you want and iterate visually. Opel is designed for maximum versatility.

**0:39** · Opel is a sandbox for rapid AI prototyping. Perfect for anyone who wants to chain models, test ideas, iterate on prompts, and experiment with tool setups before moving to production.

**0:51** · Today we're going to show a developer focused use case. How Opel lets us go from I want an app that summarizes PRs to a working UI, a chained AI workflow and a sharable tool in just a few minutes. If you're new to Opel, here's a mental model to help you throughout this demo. Opel intelligently creates multi-step workflows automatically adding steps like fetching web content, processing data, and displaying results.

### Opal’s intelligent multi-step workflows

**1:15** · In the visual workflow editor, every step is a component. Inputs, generate steps, and outputs are all visible along with their connections. This allows for fast iteration. We can change prompts, adjust steps, and rerun immediately using natural language. And with instant hosting and sharing, your app is deployed automatically and becomes sharable via a URL. We'll also get into the iteration process with Opel because prototyping isn't about getting it perfect the first time. It's about exploring and trying out ideas quickly.

**1:47** · Let's get started. Here we're starting with a blank Opal canvas where we're going to build the entire app using natural language. First, we enter our prompt.

**2:01** · Opel immediately interprets this and creates a workflow-based app. Not just a simple form, but a complete multi-step workflow. Check out how Opel automatically generates a title in an input field while intelligently creating workflow steps to fetch and process the \[music\] PR content. Okay, this is a pretty good workflow. Let's take a look.

### Exploring the generated workflow steps

**2:20** · So far, we got a user input step for the PR link, a fetch PR content step that uses the Git web page tool to retrieve the PR, a generate step to summarize the content, and an output step to display the results. Notice how each step can use different tools like maps and \[music\] web search. Let's click on the summarize PR step to see its configuration. When you select a step, Opel opens up a configuration panel on the right side.

**2:49** · Here you can see the model dropdown.

**2:52** · Opel makes it easy to choose from available models for each step.

**2:57** · You can also see the current prompt in the editing area below.

**3:02** · First, in advanced settings, we'll manually add some context about how we want the model to behave. You can edit the prompt directly, or you can use Opel's natural language editing to help update the prompt for you. We'll update it to say this. Opel has already created the output display for us as part of the automatic \[music\] workflow generation. This is useful to preview the model output while keeping in mind the model may provide different responses to the same prompt over time.

### Verifying the complete workflow

**3:30** · We'll now verify if the workflow is complete. In the editor view, we can see the GitHub PR link connected to get PR content which connects to summarize PR which connects to generate PR summary.

**3:43** · All the connections are safely made.

**3:45** · Opel's visual workflow editor shows us the complete flow. If we wanted to enhance this workflow, we could easily add more steps or tools. For example, we could upload reference documents from Google Drive using the add assets feature or add additional processing steps. But for now, our workflow is complete and ready to test.

**4:07** · Time to test our app with a real GitHub pull request.

**4:11** · If you click the app view, you'll see the immersive app experience. For now, we'll preview the app in the side panel in our editor view. In our output, we see the get clarity title and a start button. I'll click on it to begin.

**4:25** · Let's use a PR from Google's now in Android repository. Paste the link in the input field and click submit.

**4:33** · Watch as Opel runs through our chained AI workflow, fetching the PR content from the web, processing it with AI to generate a summary and displaying the result. \[music\] In seconds, we have a clear non-technical summary of the pull request. Exactly what we built this app to do. Nice. One of Opel's core \[music\] strengths is how easy it is to iterate and test changes. So, let's refine the summary prompt to improve the output. Go back to the editor view to use the suggest and edit box to refine the generate step.

### Iteration in practice: Refining output

**5:03** · Opel updates the prompt immediately. No need to redeploy or restart anything. We can test again right away to see the improved output.

**5:14** · This is the power of rapid prototyping with Opel. Iterate quickly, test immediately and refine your chained AI workflows in real time. When you're happy with your prototype, you can click the share button.

### Sharing your live prototype

**5:27** · Opel generates a public URL that you can use to share with your team. And here it is live. You can think of Opel as a place where you explore, experiment, and validate AI workflow ideas before committing to production builds. In just a few minutes, we built Git Clarity entirely through natural language and visual editing, a chain system that fetches web content, processes it with AI, and displays the results.

### Closing thoughts and future possibilities with Opal

**5:49** · Beyond Git Clarity, you could use Opel to prototype customer support bots, internal dashboards, creative tools, data processing pipelines, or any chained AI workflow you can imagine. The sky is the limit. Try Opel out at Opel.google and start building your own mini apps today.