# Kickstart Conversational Analytics agents with the Looker ChromeUX Block ## Метаданные - **Канал:** Google Cloud Tech - **YouTube:** https://www.youtube.com/watch?v=FLNw4h_-F5s ## Содержание ### [0:00](https://www.youtube.com/watch?v=FLNw4h_-F5s) Segment 1 (00:00 - 05:00) I bet you're here because you've heard about Looker's conversational analytics agents. I want to show you how easy it is to build an agent from scratch. The data we'll use today is public, meaning anyone can access it and develop their first agent. We're using the Chrome UX Looker block, which is a pre-built Look ML asset that models data from the open Chrome UX report project. With such easy access, we can go from a blank project to a live AI powered agent in about 5 minutes. The Chrome UX report isn't just practice data. It offers realworld benchmarks for websites. Look for your company's site to offer immediate impact with an AI powered agent. The data set has trends going back to 2017. You'll see how a website's performance has improved or degraded and compare one site's performance against a competitor's site. With a block, the look is already written and the metrics are defined in natural language, which means your agent has an informed knowledge engine right out of the box. So before we get started, let's make sure you have the right requirements for the project. Does your Looker have a default connection for all BigQuery projects? If not, follow the linked video for a step-by-step guide to create a connection to the Chrome UX report project. Can you access the Looker Marketplace? Talk to your admin if you don't see the shop icon on the top right corner of the Looker screen. Lastly, do you have a conversational analytics agent manager role? It's a default permission that allows you to create and manage conversational analytics agents. Now, we'll start by installing the Chrome UX block. Go to the Looker Marketplace, select discover, and search for Chrome UX. Select install and read through and agree to the terms to continue. Use the defaults or change to the connection that was set up in the requirements. Installing this block provides a model of Chrome performance data and allows you to query the data set without knowing a single line of SQL or look ML. I've included a link for a deeper understanding of the data set. Knowing your data can give you a boost when creating agents. It allows you to write detailed instructions to provide guard rails and clarify key concepts. Let's make an agent and I'll show you what I mean. Open the Looker menu and select conversations. Then select new agent. Name it something like website performance reviewer. Give it a basic description. Try something like this agent uses the Chrome UX report data set to analyze real world user experience across the web. You can ask it for performance benchmarks for your site and a competitor. Now let's connect the agent to the data set by selecting an explore. We'll choose the device summary. The country normalized explore is written to be flexible specifically for human input and isn't suitable for agents. Finally, give your agent specific instructions. Here are a few examples. First, we want the agent to stay on topic and to refocus the user if they ask questions beyond the scope of the data. When the agent is asked about LCP or sight speed, we're going to tell it to always use the P75 LCP field and to explain that this is the 75th percentile of real user experiences. We'll instruct the agent to only provide insights from January 2024 onwards. We're also going to instruct the agent to include missing URL prefixes as needed. Finally, we'll filter the responses for a default device type. You can try other instructions too. Just keep them specific and follow best practices for creating agents. Check the link for help. Let's test this agent before we publish it. Let's ensure the agent stays on topic. Okay, website performance agent, tell me what's the weather like in New York City. Okay, great. The agent triggered our custom rejection, but we did already mention weather in our prompt. So, let's try to test it with something we didn't mention. Okay, this shows us the agent seems to be grounded in our instructions, not merely its general training data, and it's not that interested in poetry. Now, let's see if it sticks to the date restrictions. Let's ask, "Show me the performance of google. com between 2021 and 2022. " See how the agent acknowledges the existence of older data, but adheres to the guardrail we imposed. Now that I've tested the agent with a few more prompts, I want to see if it delivers on the original promise of being impactful for me specifically on top of being easy to set up. We'll compare the performance of two sites. I'm ### [5:00](https://www.youtube.com/watch?v=FLNw4h_-F5s&t=300s) Segment 2 (05:00 - 05:00) going to use Google Docs and Google Scholar, but feel free to compare two sites of your own. Okay, this is awesome. It's giving me useful metrics right away. Save these agents and share them. If your users are not creating agents but using them for insights only, they'll need the conversational analytics user role. Be sure to become familiar with the limits and capabilities of AI agents and experiment with instructions. Apply clear labels, use synonyms, and address data quality to make your agents robust. Now that we've created this first agent together, explore the Chrome UX data on your own or create a new agent. Learn how to build conversational analytics agents at the link in the description. Thanks for tuning in. Chat soon. — --- *Источник: https://ekstraktznaniy.ru/video/38395*