<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Paulina Moreno on Medium]]></title>
        <description><![CDATA[Stories by Paulina Moreno on Medium]]></description>
        <link>https://medium.com/@paulinamo?source=rss-a6a8466ed041------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*IgXIj3X2GpTIKiEnULwXEw.jpeg</url>
            <title>Stories by Paulina Moreno on Medium</title>
            <link>https://medium.com/@paulinamo?source=rss-a6a8466ed041------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 16:37:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@paulinamo/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[A Noob’s Quest: Crafting Dungeons & Dragons Campaigns & Characters with Google’s Agent Development…]]></title>
            <link>https://medium.com/google-cloud/a-noobs-quest-crafting-dungeons-dragons-campaigns-characters-with-google-s-agent-development-59a3af6424f0?source=rss-a6a8466ed041------2</link>
            <guid isPermaLink="false">https://medium.com/p/59a3af6424f0</guid>
            <category><![CDATA[google-adk]]></category>
            <category><![CDATA[adk]]></category>
            <dc:creator><![CDATA[Paulina Moreno]]></dc:creator>
            <pubDate>Wed, 21 May 2025 14:11:40 GMT</pubDate>
            <atom:updated>2025-05-22T11:00:39.656Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/524/1*MAR8-ndmPKU4QHXZb_5w6w.png" /></figure><h3>A Noob’s Quest: Crafting Dungeons &amp; Dragons Campaigns &amp; Characters with Google’s Agent Development Kit (ADK).</h3><p>Ever dreamed of AI agents co-piloting your next epic <strong>Dungeons &amp; Dragons</strong> adventure? As a dev with a serious case of <strong>Shiny New Tool Syndrome</strong>, I <em>had</em> to get my hands on the <strong>Agent Development Kit (ADK)</strong> — you know, that super cool framework everyone’s buzzing about! Seeing its power for building AI agents instantly lit a wild fire in my brain: Could <strong>ADK</strong> be my magic wand to whip up a <strong>D&amp;D</strong> mini-adventure, complete with its own band of unique characters?</p><p>Welcome to my ‘Noob’s Quest with ADK’! This isn’t your typical expert guide. Instead, I’m excited to show you what I built and how I got there, sharing my live-action (okay, blog-action!) notes as I wrestled the ADK learning curve with a project I’m genuinely geeking out over. You see, while I absolutely <em>love</em> Dungeons &amp; Dragons, my brain sometimes hits a creative roadblock with adventure ideas and character backstories. And, let’s be real, I’m hardly a Dungeon Master guru! That’s exactly why ADK became my new best friend — I figured its agent-crafting magic could supercharge my world-building and character creation. Who knows, maybe with these AI sidekicks agents, my friends will <em>finally</em> let me play a proper game 😅! So, grab your dice! Let’s explore how code and creativity can roll for initiative together, proving that the most epic adventures sometimes boot up right from the command line!</p><p>Alright, before we dive deep into our D&amp;D Agent, a quick heads-up: I won’t be covering the ins and outs of ADK (like what it is or how to install it) in this post. The official documentation is excellent and has everything you need to get started. You can find it right here: <a href="https://google.github.io/adk-docs/"><em>https://google.github.io/adk-docs/.</em></a> Beyond the official docs, you’ll also find tons of great <strong>Medium</strong> posts and community articles out there that dive deep into<strong> ADK</strong>, if you’re looking to explore even further.</p><p>With that out of the way, let’s jump into the really exciting stuff: the design of our D&amp;D AI Multi-Agent!</p><p>If you’re anything like me, sometimes the ol’ imagination well runs a bit dry when it’s D&amp;D prep time. So, imagine (yeah, I know, the irony!) an assistant that helps you craft exciting Dungeons &amp; Dragons mini-campaigns, taking a huge load off your Dungeon Master shoulders. But this isn’t just about spitting out a story; it actually generates three pre-built player characters (PCs), loaded with most of the details needed for their character sheets, and even throws in a cool image for each!</p><h3>D&amp;D Multi-Agent Architecture</h3><p>So, how did we build this thing? Let’s take a peek at the blueprint. Here’s a view of the overall structure:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AUDPDK-WTGkO5sO7BAR8Dw.png" /><figcaption>Dungeons and Dragons Multi-Agent Architecture</figcaption></figure><h4><strong>Key Features Breakdown!</strong></h4><p><strong>The Root Agent</strong>: This is mission control! It’s the one that kicks off the whole campaign creation party. It hands things off to…</p><p><strong>The Story Creation Agent [Sequential Agent]</strong>: Think of this as the chief director for crafting your adventure. It orchestrates several key steps in order:</p><ol><li><strong>Iterative Story Refinement Agent [Loop Agent]</strong>: First up, this crucial loop ensures the core narrative and characters are outstanding. It cycles through the following specialist agents to progressively build and polish the story. (Little behind-the-scenes peek: I found that max_iterations=3 was the sweet spot here after some testing – it just seems to hit the right balance for quality!) Inside this loop, you’ve got:</li></ol><ul><li><strong>Writer Agent</strong>: The creative spark! This agent gets the ball rolling by writing up the main plot for your adventure.</li><li><strong>Character Writer Agent</strong>: Once the plot’s down, this agent jumps in to design unique and interesting characters that actually <em>fit</em> the story. We’re talking detailed descriptions and compelling traits. Plus, to ensure everything’s by the book (and super cool!), it taps into our D&amp;D knowledge base via Vertex AI Search, referencing official guidelines like character creation steps from the 2024 Player’s Handbook.</li><li><strong>Editor Agent</strong>: Every good story needs an editor, right? This agent provides feedback to the Writer’s narrative to make sure it’s gripping. It also double-checks that the characters from the <strong>Character Writer Agent </strong>feel like they belong in this specific tale, ensuring everything clicks.</li></ul><p>2. <strong>Penciller Agent</strong>: Once the story and characters are refined by the loop above, the Penciller Agent steps in. It takes the character concepts and generates that cool artwork you’ll see for each PC, really helping to bring them to life. So yeah, you can finally stop just <em>imagining</em> what they look like!</p><p>3. <strong>Publisher Agent</strong>: With all the narrative, character details, and awesome artwork ready, the Publisher Agent takes over. It neatly packages everything into a final HTML page, bundling up the whole adventure — story, character info, and art — all ready for your next game!</p><h4><strong>The Toolkit.</strong></h4><p>Our D&amp;D Multi-Agent doesn’t conjure campaigns out of thin air (as cool as that would be!). As I mentioned before, it uses some powerful tools to make the magic happen. Here’s a look at a couple of key integrations:</p><p><strong>Vertex AI Search Integration — </strong>To make this integration effective, a Vertex AI Search Datastore first needs to be created and populated with the respective D&amp;D content. Our agent use <em>Vertex AI Search</em> to:</p><ul><li><strong>[characters_vertex_search_tool] </strong>— <strong>Nail the Rules for Characters:</strong> Quickly access and understand the comprehensive rules from the D&amp;D 2024 Player’s Handbook. This is crucial for the <strong>Character Creator Agent</strong> to ensure all those pre-made PCs are legit and follow the latest guidelines.</li><li><strong>[campaign_vertex_search_tool] — Spark Campaign Ideas:</strong> Provide access to a wealth of campaign examples. This can give the <strong>Writer Agent</strong> a springboard for inspiration or help it structure new adventures.</li></ul><p><strong>DALL-E Integration (via CrewAI) — The Artist’s Touch:</strong> You know those cool character images we mentioned? This is how they come to life!</p><ul><li><strong>[adk_dalle_tool]</strong> — <strong>Visualizing Your Heroes:</strong> The <strong>Penciller Agent</strong> uses DALL-E model, hooked up through the CrewAI framework, to generate unique character images. It takes the textual descriptions and — <em>voila!</em> — DALL-E helps create a visual representation.</li></ul><h3><strong>D&amp;D Agent in Action</strong></h3><p>So, what does this look like when you fire it up? Let’s walk through a typical run with adk web command to interact in the ADK Web UI<strong>.</strong></p><p>First things first, you’ll need to kick off the interaction. (Quick tech note: for this version, a simple “Hello” works for this version– a little quirk I’m planning to smooth out in a future update!).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s-_bRr0bHzDE-JnSDPUzfA.png" /><figcaption>D&amp;D — Root Agent interaction</figcaption></figure><p>Once you’ve said hi, the Agent greets you and asks for the main theme or topic of your mini-campaign. This is where you plant the creative seed!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uip2rdQIeC4p4GAzXAR1lQ.png" /><figcaption>D&amp;D — Campaign Theme</figcaption></figure><p>From that point on, the <strong>D&amp;D multi-agent</strong> takes the reins and works its magic. It gets down to business:</p><ul><li><strong>Crafting the Story [Writer Agent]:</strong> It writes the core narrative for your mini-campaign.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SIYU4ivUWCAbv3mF_8D6pQ.png" /><figcaption>Writer Agent — First Draft</figcaption></figure><pre>writer_agent = Agent(<br>    name=&quot;writer_agent&quot;,<br>    model=os.getenv(&quot;MODEL_NAME&quot;),<br>    description=&quot; Writes the initial draft of the Dungeons and Dragons mini campaign adventure. &quot;,<br>    instruction=WRITER_PROMPT,<br>    tools=[campaign_vertex_search_tool],<br>    output_key=&quot;current_story&quot;,<br>)</pre><pre>WRITER_PROMPT = &quot;&quot;&quot;<br>**Role:** You are an expert Dungeon Master and creative writer.<br><br>**Goal:** Generate a three-chapter Dungeons and Dragons mini-campaign adventure for level 1 to 5 based on the provided topic.<br><br>**Context:**<br>- User Prompt: {{ prompt }}<br>- Editor Feedback: {{ editor_feedback? }}<br><br>**Instructions:**<br>1.  **Create a Mini-Campaign:** Develop a compelling three-chapter D&amp;D adventure based on the User Prompt.<br>2.  **Incorporate Feedback:** If Editor Feedback is provided, revise the story accordingly to improve it.<br>3.  **Use Tools:** Utilize the &#39;campaign_vertex_search_tool&#39; to gather information on creating mini-campaigns if needed.<br>4.  **Required Content:** The adventure book must include:<br>    *   A guide for the Dungeon Master (DM) to run the campaign.<br>    *   Campaign can only be for 3 players maximum.<br>    *   Chapter-by-chapter breakdown with clear objectives, major events, and potential player choices.<br>    *   Non-Player Character (NPC) profiles detailing personality traits, goals, and roleplaying tips.<br><br>**Output Format:**<br>- Provide *only* the final adventure book chapters as the output.<br>- Do not include any introductory text, preamble, or explanations outside the adventure content itself.<br>&quot;&quot;&quot;</pre><ul><li><strong>Creating Characters[Character Writer Agent]:</strong> Unique player characters that fit the story.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wxNN7DUjycmZuUui44qQuA.png" /><figcaption>Mini Campaign — D&amp;D Characters Description</figcaption></figure><p>Here’s a look at a small part of the prompt :</p><pre>CHARACTER_CREATOR_PROMPT = &quot;&quot;&quot;<br>**Role:** You are an expert character creator for Dungeons and Dragons.<br><br>**Goal:** Create **three** distinct **level 1** player characters suitable for the provided D&amp;D mini-campaign story.<br><br>**Context:**<br>- Current Story: {{ current_story }}<br>- Editor Feedback: {{ editor_feedback? }}<br><br>**Instructions:**<br>1.  **Analyze Context:** Review the `Current Story` and any `Editor Feedback` provided.<br>2.  **Create Characters:** Generate three unique *level 1* player characters that fit the story&#39;s themes and potential challenges.<br>3.  **Character Details (For each character):** Using the tool `characters_vertex_search_tool` for D&amp;D rules and guidelines, define the following for each character:<br>        * Choose a suitable Class: The class define what the character is capable of doing and what role they play in the game. Each class provides:<br>            - Hit points, which determine how much damage the character can take.<br>            - Proficiencies, which include skills, weapons, armor, and tools the character is good with.<br>            - Class features, which are unique abilities that develop as the character levels up.<br>        * Determine an appropriate Origin: The character&#39;s origin consists of their species and background:<br>            - Species: Describes the character&#39;s biological and cultural traits (e.g., Elf, Human, Dragonborn). Each species offers traits such as darkvision, innate magic, or movement abilities.<br>            - Background: Represents the character&#39;s life experience before adventuring (e.g., Acolyte, Soldier, Artisan). Backgrounds grant:<br>                - Skill proficiencies<br>                - Starting equipment<br>                - A unique feature that helps shape the character&#39;s identity and roleplay<br>...</pre><ul><li><strong>Refining &amp; Iterating[LoopAgent]:</strong> Remember that Iterative Story Refinement Loop? This is where it ensures the story and characters are compelling, making adjustments based on its internal checks.</li></ul><pre># Code from main agent.py<br>iterative_story_refinement_agent = LoopAgent(<br>    name=&quot;iterative_story_refinement_agent&quot;,<br>    description=&quot;Iterates through writing and editing to improve the D&amp;D mini campaign adventure.&quot;,<br>    sub_agents=[<br>        writer_agent,<br>        character_creator_agent,<br>        editor_agent<br>    ],<br>    max_iterations=3<br>)</pre><pre># Code from ./sub_agent/editor/agent.py<br>editor_agent = Agent(<br>    name=&quot;editor_agent&quot;,<br>    model=os.getenv(&quot;MODEL_NAME&quot;),<br>    description=&quot;&quot;&quot;<br>    Enhances the overall quality, clarity, consistency, and professional polish of all content in the D&amp;D mini campaign adventure.<br>    &quot;&quot;&quot;,<br>    instruction=EDITOR_PROMPT,<br>    output_key=&quot;editor_feedback&quot;,<br>    tools=[exit_loop],<br>)</pre><pre>EDITOR_PROMPT = &quot;&quot;&quot;<br>**Role:** You are a highly experienced Dungeons &amp; Dragons campaign editor and content evaluator, specializing in refining narrative for optimal gameplay.<br><br>**Goal:** To meticulously review and enhance the provided D&amp;D campaign story and character descriptions, ensuring they are narratively compelling, consistent, clear, easy for a Dungeon Master to run, and adhere to D&amp;D conventions.<br>Your task is to deliver actionable, focused feedback for improvement or signify approval.<br><br>**Context:**<br>- **Campaign Narrative:** {{ current_story }}<br>- **Key Characters (Players):** {{ characters_brief }}<br><br>**Instructions:**<br>Critically evaluate the `Campaign Narrative` alongside the `Key Characters` based on the following D&amp;D-centric criteria. Provide precise, constructive feedback.<br><br>1.  **Editorial Polish:** Check for errors in grammar, syntax, spelling, and punctuation. Ensure consistent style and tone, and improve overall clarity, conciseness, and readability for a Dungeon Master.<br>2.  **D&amp;D Terminology &amp; Lore Fidelity:** Verify accurate and consistent use of D&amp;D terms (e.g., &quot;AC,&quot; &quot;CR,&quot; &quot;Saving Throw&quot;). Ensure names (characters, places, items) are consistent and that the narrative respects established D&amp;D lore or internal campaign lore.<br>3.  **Narrative Structure &amp; Pacing:** Assess the story&#39;s flow, pacing, and tension development. Does it have clear hooks, rising action, and a logical progression? Suggest improvements for engagement and coherence.<br>4.  **Internal Consistency &amp; World Logic:** Identify and flag any contradictions in plot, character motivations, or world details. Ensure the narrative logic holds together from start to finish.<br>5.  **Character Depth &amp; Integration (Players):** Evaluate if characters are well-developed, have clear motivations, and effectively serve their narrative purpose. Are they integrated smoothly into the story, and can a DM easily roleplay them?<br>6.  **DM Usability &amp; Player Agency:** Is the content easy for a Dungeon Master to prepare and run? Does the narrative provide meaningful choices and agency for the players, avoiding excessive railroading?<br>7.  **Challenge &amp; Balance (if present):** If encounters or challenges are detailed, are they appropriate for the intended level range? Are the mechanics clearly described for the DM?<br><br>**Output Format:**<br>* **Approved:** If the content is polished, compelling, and fully ready, YOU MUST call the `exit_loop` function.<br>* **Feedback:** If revisions are necessary, provide detailed, specific, and actionable feedback. For each point, clearly:<br>    * State the **Criterion** it relates to (e.g., &quot;Narrative Structure&quot;).<br>    * Describe the **Issue** (e.g., &quot;The climax feels rushed and lacks sufficient build-up.&quot;).<br>    * Explain the **Impact** (e.g., &quot;This might leave players feeling unsatisfied with the campaign&#39;s resolution.&quot;).<br>    * Suggest the **Area for Improvement** (e.g., &quot;Consider adding another significant challenge or a moral dilemma before the final confrontation to heighten stakes.&quot;).<br>    * *Do not rewrite the story.* Focus solely on identifying and explaining areas for improvement.<br>&quot;&quot;&quot;</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ih-Qj3-DpJPDLZGSCXjT5A.png" /><figcaption>Editor Agent Feedback</figcaption></figure><p>The <strong>Writer Agent</strong> proceeds by taking the <strong>Editor’s feedback</strong> and iterating on the current story to improve it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lItZwWw8Zpc4A-dE7hJy6g.png" /><figcaption>Writer Agent — Second Draft</figcaption></figure><ul><li><strong>Generating Artwork [Penciller Agent]:</strong> Once the content is polished, the Penciller Agent creates unique artwork for each character using CrewAI’s <a href="https://docs.crewai.com/tools/dalletool">DALL-E tool</a></li></ul><pre># Code from main agent.py<br>story_creation_agent = SequentialAgent(<br>    name=&quot;story_creation_agent&quot;,<br>    description=&quot;The story team agent is responsible for creating the story for the D&amp;D mini campaign adventure.&quot;,<br>    sub_agents=[<br>        iterative_story_refinement_agent,<br>        penciller_agent,<br>        publisher_agent<br>    ]<br>)<br></pre><pre># Code from ./sub_agent/penciller/agent.py<br>penciller_agent = Agent(<br>    name=&quot;penciller_agent&quot;,<br>    model=os.getenv(&quot;MODEL_NAME&quot;),<br>    description=&quot;&quot;&quot;<br>    Generates fantasy character images in a style suitable for Dungeons and Dragons mini campaign adventure illustrations, using DALL-E.<br>    &quot;&quot;&quot;,<br>    instruction=PENCILER_PROMPT,<br>    tools=[adk_dalle_tool],<br>    output_key=&quot;character_image_urls&quot;<br>)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1DkqFFY0lixdFq8BfftDSg.png" /><figcaption>Penciller Agent- DALL-E characters generated image</figcaption></figure><ul><li><strong>Publishing[Publisher Agent]:</strong> Finally, it bundles everything–the story, character details, and images–into a neat HTML file.</li></ul><pre># Code from ./sub_agent/publisher/agent.py<br>publisher_agent = Agent(<br>    name=&quot;publisher_agent&quot;,<br>    model=&quot;gemini-2.5-pro-preview-03-25&quot;,<br>    description=&quot;Execute the code to publish the Dungeons and Dragons mini campaign adventure to a website&quot;,<br>    instruction=PUBLISHER_PROMPT,<br>    output_key=&quot;publication_html&quot;<br>)</pre><p>Okay, so getting the <strong>Publisher Agent</strong> to create well-formatted HTML was a bit of a fun challenge at first! Trying to get the <em>exact</em> HTML structure I wanted the agent to output on its own felt like wrestling a particularly stubborn bit of code. So, I decided to give it a helping hand with an HTML template. Seriously, this made a <em>huge</em> difference in getting things to look just right.</p><pre>PUBLISHER_PROMPT = &quot;&quot;&quot;<br>**Objective:** Generate a single, complete, and visually appealing responsive HTML page for a Dungeons &amp; Dragons mini-campaign, effectively presenting the campaign story and character information.<br><br>**Context Variables:**<br>* - Current Story: {{ current_story }} (String) Markdown formatted campaign story.<br>* - Characters Description: {{ characters_brief }} (JSON) Detailed information for each character. <br>* - Characters Images: {{ character_image_urls }} (JSON) URLs for character images, corresponding to the each character in `{{ characters_brief }}`.<br><br>**Core Requirements:**<br>1.  **HTML Structure &amp; Semantics:**<br>    * Generate a single, valid HTML5 document.<br>    * The page title (`&lt;title&gt;`) should be the title of the campaign, derived from `Current Story`.<br>    * Ensure the page is responsive to different screen sizes.<br><br>2.  **Styling:**<br>    * Generate D&amp;D-themed CSS. Aim for a look and feel reminiscent of classic D&amp;D books (e.g., parchment or dark textured backgrounds, thematic borders, potentially scroll-like elements).<br>        * Use &quot;Cinzel Decorative&quot; font for main headings (h1, h2) and &quot;EB Garamond&quot; for body text. Ensure these fonts are imported (e.g., via Google Fonts) if not standard.<br>        * Implement clear visual separation between major sections (e.g., using borders, background changes, or distinct spacing).<br>    * Ensure proper spacing, padding, and margins for readability.<br><br>3.  **Campaign Story Section:**<br>    * This section should appear first under the main page heading.<br>    * Convert the Markdown content from `Current Story` into well-formed HTML.<br>    * Use appropriate heading hierarchy (e.g., `&lt;h1&gt;` for the main campaign title derived from `Current Story`, `&lt;h2&gt;` for major story sections, `&lt;h3&gt;` for sub-sections, etc.).<br>    * Format paragraphs (`&lt;p&gt;`) and other text elements (bold, italics) correctly.<br><br>4.  **Character Sections:**<br>    * Create a distinct section for each character detailed in `Characters Description`.<br>    * **Layout:** Arrange character sections in a consistent and visually appealing manner (e.g., using flexbox or grid for a gallery if multiple characters, or stacked sections).<br>    * **Content per Character:**<br>        * **Image:** Display the character&#39;s image using the corresponding URL from `Characters Images`.<br>            * Include descriptive `alt` text for the image (e.g., &quot;Image of [Character Name]&quot;).<br>            * If an image fails to load, ensure a simple text placeholder like &quot;[Character Name - Image Unavailable]&quot; is visible, or a styled placeholder box.<br>        * **Information Structure:** Present information in a clear, organized format (e.g., using definition lists, styled divs, or sub-sections):<br>            * Character Name (as a prominent heading for the character&#39;s section)<br>            * Character Quote (if available)<br>            * Background and Personality<br>            * Features and Abilities<br>            * Character Stats and Attributes<br>            * Equipment and Actions<br>        * **Styling:** Apply a consistent &quot;character sheet&quot; feel to each character&#39;s presentation<br><br> //... prompt continues ...<br></pre><p>Now, for the agent’s model choice, I ended up switching to gemini-2.5-pro. I first gave flash a shot, and even though I provided the HTML template as a guide, flash didn’t consistently deliver the structured output I was after. Switching to gemini-2.5-pro did the trick, getting me those results I needed!</p><p>And <em>boom</em>! Here’s what the final HTML looks like with everything pulled together:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DOH4zDbrd-I2QPEzWAaLqw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xOQcxQDv59Oh8VPgW22tIA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lQaPME5unc7TFfQ7f3zxkw.png" /><figcaption>Mini Campaign HTML section</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1ltg_GYb6kHsAvx5DaYhIw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dclO_H6_4dFKPrmRLMiTRQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CQahahohXa1yBIjDBB-vpA.png" /><figcaption>D&amp;D Character Sheets</figcaption></figure><p>I know this could definitely look even better — and as I’ve mentioned before<strong>, </strong>I’m not a D&amp;D expert by any means<strong>,</strong> so some of the character stats might be a bit off — but hey, I’m pretty happy with it for now! I’ll be sure to polish it up more down the line when I find a spare moment (or two!).</p><p>This multi-agent system is really just scratching the surface, and my ideas list for what’s next is already buzzing! I’m currently working on some cool under-the-hood improvements, like adding <strong>before/after</strong> callbacks to better handle things like guardrails and ensure consistent output formatting.</p><p>And that’s just for starters! Down the road, I’m envisioning even bigger possibilities — imagine a dedicated web app making your <strong>D&amp;D</strong> prep even smoother, or a handy phone app for generating quick adventure ideas on the go. There’s a huge world of potential here, and I’m pumped to keep evolving this project, diving deeper into <strong>ADK</strong>, and exploring all the awesome related tech. I hope this post hit the mark for fellow <strong>ADK</strong> explorers and <strong>D&amp;D</strong> enthusiasts alike 🤖</p><p><strong>Links</strong>:</p><ul><li><strong>GitHub Repository:</strong> <a href="https://github.com/paulinaMorenoA/DnD-adk/tree/main">here</a>.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=59a3af6424f0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/google-cloud/a-noobs-quest-crafting-dungeons-dragons-campaigns-characters-with-google-s-agent-development-59a3af6424f0">A Noob’s Quest: Crafting Dungeons &amp; Dragons Campaigns &amp; Characters with Google’s Agent Development…</a> was originally published in <a href="https://medium.com/google-cloud">Google Cloud - Community</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google Agent Assist: Redefining Customer Service Through AI-Powered Efficiency]]></title>
            <link>https://medium.com/google-cloud/google-agent-assist-redefining-customer-service-through-ai-powered-efficiency-cd0bd5217b9b?source=rss-a6a8466ed041------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd0bd5217b9b</guid>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[agent-assistant]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Paulina Moreno]]></dc:creator>
            <pubDate>Mon, 07 Oct 2024 07:06:43 GMT</pubDate>
            <atom:updated>2024-10-07T21:07:17.443Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c0gN_Pc0y50iBYojpNbXFg.jpeg" /></figure><p>In the fast-paced world of technology, customer service has taken on a new significance, becoming an essential part of brand experience and customer loyalty. Meeting customer demands means providing immediate answers and personalized support. Ideally, you’d be connected to a well-informed agent right away, who can swiftly and effectively address any concerns.</p><p>By now, you’ve likely heard about Google Cloud’s Contact Center AI (CCAI). This powerful suite of tools is designed to revolutionize contact centers, balancing exceptional customer experiences and operational efficiency. The CCAI suite includes:</p><ul><li><strong>Virtual agent powered by Dialogflow:</strong> A tool for creating advanced virtual agents that handle routine tasks and interactions 24/7.</li><li><strong>Contact Center AI Insights:</strong> This product uses natural language processing to help managers identify call drivers, customer sentiment, popular questions, and other key details.</li><li><strong>Agent Assist:</strong> An AI-powered tool that understands customer intent and provides real-time guidance to employees during calls and chats.</li></ul><p>For the past few months, I’ve been working closely with Agent Assist, and I’m excited to share more about this innovative tool. It enhances agent performance, increases efficiency, and reshapes how customer support is delivered. Although it’s not brand new, Agent Assist continues to develop, helping businesses improve their customer support in significant ways.</p><h3>A Deeper Look: What Makes Google Agent Assist Stand Out</h3><p>Imagine a customer service agent who never has to pause awkwardly to search for information. With Google Agent Assist, this is a reality. It offers real-time, contextually relevant suggestions, transforming how agents interact with customers. It’s not only about helping with regular tasks or questions, but it’s also about taking initiative, reducing the workload of agents, and improving the overall customer experience. But what makes this tool stand out is its ability to learn from your best agents and adapt to the specific needs of a business.</p><p>Through the analysis of previous interactions, Agent Assist can identify patterns, detect frequently encountered issues, and continuously improve its suggestions based on these insights.</p><p>To understand why Google Agent Assist stands out, let’s examine some of its key elements.</p><p><strong>Contextual and Real-Time Assistance</strong>: One of the primary strengths of Agent Assist lies in its ability to provide on-the-fly support, delivering suggestions and relevant information in real time. Whether it’s recommending articles from the knowledge base or suggesting the best way to resolve a customer issue, the tool ensures agents can focus on building relationships with customers rather than searching for information.</p><p><strong>Customization for Unique Business Needs</strong>: No two businesses are alike, and customer service teams often have unique workflows, communication styles, and brand requirements. Agent Assist customizes its responses to match a company’s brand voice and approach by analyzing the performance of its best agents.</p><h3>Agent Assist Integrations</h3><p>Agent Assist supports agents text and voice channels.</p><p><strong><em>Note: The diagrams provided illustrate only a high-level overview and do not represent the complete architecture.</em></strong></p><h4>1. Chat Integration Process</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w-Z6OzXK4fLCc8Zct1GHnQ.jpeg" /><figcaption>AA Chat Integration</figcaption></figure><p>When a customer initiates a chat, the request is processed via <strong>Dialogflow APIs</strong> that are connected to the business’s customer chat backend. Agent Assist analyzes the conversation, providing real-time transcripts and delivering actionable suggestions to the agent’s desktop interface.</p><p>To obtain Agent Assist suggestions, call the Dialogflow <a href="https://cloud.google.com/dialogflow/es/docs/reference/rest/v2/projects.conversations.participants/analyzeContent">AnalyzeContent</a> API with the new message. The response body will contain an instance of <a href="https://cloud.google.com/dialogflow/es/docs/reference/rest/v2beta1/AnalyzeContentResponse">AnalyzeContentResponse</a> with the the suggestions for most recent human agent under <strong>humanAgentSuggestionResults </strong>field.</p><h4>2. Voice Integration Process</h4><p>Voice integration in Agent Assist leverages <strong>gRPC-based APIs</strong> and <strong>SIPREC</strong> (Session Initiation Protocol Recording). While both methods achieve similar outcomes, they offer different technical approaches:</p><p><strong>SIPREC-based Integration</strong>: <a href="https://medium.com/google-cloud/siprec-integration-between-twilio-flex-and-ccai-agent-assist-6ce46d1a2ca2"><strong>SIPREC</strong></a> stands for Session Recording Protocol. Its primary function is to enable call recording in telephony environments. It works as an extension to Session Initiation Protocol, or SIP. SIPREC works in combination with <a href="https://cloud.google.com/speech-to-text/docs"><strong>Cloud Speech-to-Text</strong></a> to generate transcripts of your audio data before Agent Assist can use it to generate suggestions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1xzF-9WzOr-48yD2abu5Tg.jpeg" /><figcaption>AA- SIPREC</figcaption></figure><p>With SIPREC integration the customer audio data is streamed to CCAI and Agent Assist in two media, one for the end user and the other for the human agent. Agent Assist can listen, comprehend the conversation, and provide useful advice or suggestions to the agent, causing no interruptions. These suggestions are delivered to the agent’s desktop application either through REST APIs or by subscribing to the relevant Pub/Sub topic.</p><p><strong>gRPC API Integration : </strong>Using Google’s <strong>Dialogflow </strong><a href="https://cloud.google.com/dialogflow/es/docs/reference/rpc/google.cloud.dialogflow.v2beta1#google.cloud.dialogflow.v2beta1.Participants.StreamingAnalyzeContent"><strong>StreamingAnalyzeContent</strong></a> method, businesses can stream real-time audio from customer interactions, providing transcriptions that are fed back to agents along with suggested responses. This method is great for real-time streaming and analysis, but it might have a bit more delay. Additionally, you’ll need to manage the separation of the conversation into individual participants before sending it to Dialogflow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oRqTk57zR4GM_4Uv_WTRGg.jpeg" /><figcaption>AA-gRPC Integration</figcaption></figure><p>To send audio from the relevant participant, you must initiate a call to the dialogflow <strong>StreamingAnalyzeContent </strong>method<strong>.</strong> Simultaneously, this will provide real-time transcriptions. StreamingAnalyzeContent recives a<strong> InputAudioConfig</strong> parameter to instruct the speech recognizer on how to process the audio content, like the <a href="https://cloud.google.com/speech-to-text/docs/basics#select-model">Speech-to-Text model</a> that you want to used to transcribe your audio.</p><h3><strong>Explore </strong>Agent Assist Features</h3><p>Here’s a list of the current Agent Assist features. While some of them are still in preview, I think they’re definitely worth mentioning!</p><ul><li>1. <strong>Smart Reply</strong>[💬]<br><em>Streamlines chat interactions by suggesting accurate, context-specific replies, saving time for agents and customers.</em></li><li>2. <strong>Generative Knowledge Assist</strong>[📚]<br><em>Answers agents’ questions by referencing information from provided documents.</em></li><li>3. <strong>Proactive Generative Knowledge Assist</strong>[🔍] + [📚]<br><em>Follows ongoing conversations and proactively offers search suggestions based on the dialogue.</em></li><li>4. <strong>Baseline LLM Summarization</strong>[📝]<br><em>Generates real-time call summaries with customizable styles and focus areas.</em></li><li>5. <strong>Sentiment Analysis</strong>[😊😠]<br><em>Detects customer emotions during interactions, revealing both positive and negative sentiments.</em></li><li>6. <strong>FAQ Assist [Preview]</strong>[❓]<br><em>Suggests relevant FAQ answers to agents during live conversations to help with common customer queries.</em></li><li>7. <strong>Article Suggestion</strong>[📄]<br><em>Suggests relevant documents for agents to review or share during customer interactions.</em></li><li>8. <strong>Live Transcription</strong>[🎤]<br><em>Transcribes conversations in real-time between the agent and the customer.</em></li></ul><h3><strong>Agent Assist — Conversation Profile</strong></h3><p>Whether you’re using voice or text integration, the first step is to create your Conversation Profile (You must create a conversation profile in order to create a conversation with an end-user). Think of this profile as your personal dashboard for managing all the Agent Assist features. You can customize your <a href="https://cloud.google.com/agent-assist/docs/conversation-profile">Configuration Profile</a> with the specific suggestions that will help you the most.</p><p>Setting up your Conversation Profile is easy — you can do it through the <a href="https://agentassist.cloud.google.com/">Agent Assist Console</a> or by using the <a href="https://cloud.google.com/agent-assist/docs/how-to">API directly</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KO0FoxGIYnAxNsGSk5MmDw.png" /><figcaption>Conversation Profile</figcaption></figure><p>But it doesn’t stop there! You also have the power to manage security and <a href="https://cloud.google.com/agent-assist/docs/transcription#send_audio_data_and_get_a_transcript">speech-to-text settings</a>. These features are essential for handling data redaction and retention. For example, they help determine how information is redacted in cloud logging and interaction logs, keeping your conversation history secure. So take a moment to tailor these settings to fit your needs!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*05lu_bMzyfBr6TmaBp80Fw.png" /><figcaption>Configuration Profile — Security Settings</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/826/1*qhwvbPLpJFdEpwlCbrEY0Q.png" /><figcaption>Conversation Profile — STT configuration</figcaption></figure><h3>It’s All About the Feedback: How to Keep Agent Assist in Check</h3><h4>Give Your Two Cents to Agent Assist</h4><p>If you ever feel like Agent Assist needs some guidance, here’s how you can provide feedback on the suggested answers that are not quite perfect.</p><ol><li><strong>Implicit Feedback: </strong><br>Whether or not an agent clicked on a suggested answer. This information is contained in the boolean clicked field.</li><li><strong>Explicit Feedback: The Truth Serum</strong><br>How accurate was that answer? Rate its correctness and spill any extra details in the agentAssistDetailFeedback field using the AgentAssistantFeedback object.</li></ol><h4>How to Send Feedback:</h4><p>Every response from Agent Assist comes with an Answers field, like faqAnswers or articleAnswers. To share your thoughts:</p><ul><li><strong>Track Down the Record</strong>: Find the answer’s resource name in the format “projects/project-id/answerRecords/answer-record-id”.</li><li><strong>Make It Official</strong>: Use this record to call the UpdateAnswerRecord method and submit your AnswerFeedback.</li></ul><p><strong>Feedback Example</strong>:</p><p>The correctnessLevel field can be one of FULLY_CORRECT, PARTIALLY_CORRECT, NOT_CORRECT, CORRECTNESS_LEVEL_UNSPECIFIED.</p><pre>curl -X PATCH \<br>     -H &quot;Autorization: Bearer $(gcloud auth application-default print-access-token)&quot; \<br>     -H &quot;Content-Type: application/json&quot; -d<br>&#39;{<br>  &quot;name&quot;: &quot;projects/project-id/answerRecords/answer-record-id&quot;,<br>  &quot;answerFeedback&quot;: {<br>    &quot;correctnessLevel&quot;: &quot;FULLY_CORRECT&quot;,<br>    &quot;agentAssistantDetailFeedback&quot;: {},<br>    &quot;clicked&quot;: true<br>  }<br>}&#39; https://dialogflow.googleapis.com/v2/projects/project-id/answerRecords/xxxxx?updateMask=answer_feedback</pre><h3>Agent Assist- <strong>Transforming Customer Service Across Industries</strong></h3><p>The potential applications of Agent Assist are far-reaching, particularly when viewed through the lens of various industries. From healthcare to finance, businesses can deploy Agent Assist to tackle customer service challenges in unique ways. Some potential use cases include:</p><ul><li><strong>Finance</strong>: Offering contextually relevant advice during sensitive financial transactions or loan applications, ensuring agents remain compliant with regulations while giving customers accurate and personalized responses.</li><li><strong>Retail</strong>: Assisting agents with product recommendations, troubleshooting, and order tracking, ultimately leading to faster resolution times and improved customer satisfaction.</li></ul><h3>Empowering Human Agents Through AI</h3><p>Google Agent Assist is a transformative tool designed to enhance, not replace, the invaluable role of human agents. By offering real-time insights, tailored suggestions, and strategic guidance, it empowers agents to elevate customer experiences to new heights. For businesses seeking to streamline their customer service operations while fostering deeper, more meaningful interactions, Agent Assist is a revolutionary ally — one that promises to reshape the future of customer service while honoring the essential human touch.</p><p>🌵 <strong>Until Next Time, Y’all!</strong> 🤠</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd0bd5217b9b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/google-cloud/google-agent-assist-redefining-customer-service-through-ai-powered-efficiency-cd0bd5217b9b">Google Agent Assist: Redefining Customer Service Through AI-Powered Efficiency</a> was originally published in <a href="https://medium.com/google-cloud">Google Cloud - Community</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploring Google Vertex AI Conversation — Dialogflow CX Generative AI features Data store &…]]></title>
            <link>https://medium.com/google-cloud/exploring-google-vertex-ai-conversation-dialogflow-cx-generative-ai-features-data-store-34d8813d3875?source=rss-a6a8466ed041------2</link>
            <guid isPermaLink="false">https://medium.com/p/34d8813d3875</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[conversational-ai]]></category>
            <category><![CDATA[dialogflow-cx]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[coversational-bots]]></category>
            <dc:creator><![CDATA[Paulina Moreno]]></dc:creator>
            <pubDate>Mon, 26 Feb 2024 04:37:05 GMT</pubDate>
            <atom:updated>2024-02-26T06:25:08.034Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yzlTglgJDBBjwqIIAyzP7w.png" /></figure><h3>Exploring Google Vertex AI Conversation — Dialogflow CX Generative AI features Data store &amp; Generators</h3><p>Vertex AI Conversation comprises generative conversational features developed on Dialogflow and Vertex AI. These capabilities empower the utilization of large language models (LLMs) for parsing and understanding content, crafting agent responses, and managing the flow of conversations. This enhancement can minimize the time required for agent design and enhance the overall quality of agents.</p><p>With the introduction of Vertex AI Conversation now you can leverage a state machine approach to create sophisticated agents with generative AI capabilities for dynamic conversation design and automation.</p><p>In this blog post, we take a close look at the innovative technology of Dialogflow CX’s Generative AI. We will be focus on <strong>Data stores </strong>and <strong>Generators</strong>.</p><p>Think of <strong>Data Stores</strong> as a huge library of information. When you ask a question, a virtual assistant (like a super-smart librarian) searches this library for the answer. <strong>Dialogflow CX</strong> has a special tool called “Data Store” that makes it easy to create conversations focused on information from your library.</p><p>We can integrate various sources into a Data store, including:</p><ul><li><strong>Website domain</strong> — Google offers to index your website domain, but you’ll need to go through a verification process to confirm ownership.</li><li><strong>Unstructured data </strong>— Private organizational information (in formats like HTML, PDF, and TXT files)(<strong>Note</strong>: PPTX and DOCX formats are available in Preview).</li><li><strong>Structured data</strong> — Preview(BigQuery or Cloud Storage)</li></ul><p>Keep in mind that there are limitations on file size and the number of files. To learn more about preparing your data, visit: <a href="https://cloud.google.com/generative-ai-app-builder/docs/prepare-data">https://cloud.google.com/generative-ai-app-builder/docs/prepare-data</a>.</p><h3>Dialogflow CX Generators</h3><p>By using generators, you can summon a large language model (LLM) inline within Dialogflow CX with no external webhook. The Generator can be customized to perform a range of tasks, including summarization, extracting parameters, and manipulating data. Generators are sourced from Vertex AI, delivering dynamic agent behavior and real-time responses based on the prompts you provide.</p><p>The GCP public documentation offers a compilation of commonly used <a href="https://cloud.google.com/dialogflow/cx/docs/concept/generators#examples">Generator recipes</a>. While these recipes have proven to work well, it is important to use them cautiously, as we are using a LLM and its predictability can vary.</p><p>To demonstrate text summarization, we’ll be using generators a bit later in the example. Let’s dive in!</p><h4>Create a Chat Application</h4><p>In the Google Cloud console, go to the Search and Conversation page.</p><figure><img alt="GCP Search &amp; conversation page" src="https://cdn-images-1.medium.com/max/1024/0*bYYdJpOsvpsIZnFA" /><figcaption>GCP Search &amp; Conversation page</figcaption></figure><p>When you visit Search and Conversation for the first time, you can expect to see the following page:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tY40oWEodu6PmZLV" /><figcaption>Search &amp; Conversation welcome page</figcaption></figure><p>Read and agree to the Terms of Service, then click Continue and activate the API</p><p>Next select the “<strong>Chat</strong>” option</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EhAHCW5BjYQUvjHKNBhCxg.png" /><figcaption>Application Type</figcaption></figure><p>Add the name of your company(it doesn’t have to be an actual company) and the name of the agent</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/926/1*ACqsQEbVZOsq75-YRQmzEQ.png" /><figcaption>Agent Configurations</figcaption></figure><p>The next setup will involve data. Our example will make use of the Cloud Storage option. I have put together a PDF that covers everything you should know about Shiba Inu dogs.</p><p><strong><em>Note: </em></strong><em>You see, I’m a Shiba owner, and these fluffy drama queens have completely stolen my heart</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*L2HfopZNjT4T4Uag" /><figcaption>Data Source</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hhT5HkkoPasy92hO" /><figcaption>GCS bucket to import</figcaption></figure><p>Next we have additional data store configurations</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MC2A8QJArcF4cvXm" /><figcaption>Data store name</figcaption></figure><p>Now you can see our new chat application under <strong>Search &amp; Conversation</strong> | <strong>Apps</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wnW03jC8fKAlQo1UVWBbxw.png" /></figure><p>Select your app and the link will take you to Default Start page on <strong>Dialogflow CX</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vTe22qQmw4FEem4r" /><figcaption>Dialogflow CX -Agent Start Page</figcaption></figure><p>On the Start page, navigate to the <strong>Data stores</strong> section to locate the name of the data store we previously created for our bot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YFEH1G988rGlX1B5tlom0A.png" /><figcaption>DialogFlow CX — Data stores configuration</figcaption></figure><p>Now we can start asking questions to our agent. Click on <strong>Test Agent </strong>to open the simulator, leave the parameters (environment, flow, page) as default.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/424/1*RzlK5mOSYArgK0DiL2pCvg.png" /><figcaption>Agent simulator</figcaption></figure><p>On the Simulator window, click the clipboard icon:</p><p>This will open a pop-up window with the o<strong>riginal response (JSON) </strong>from the agent</p><p>This JSON object has many details about the process steps and the provided response. We’ll focus on how the response is based on factual information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cZ1WVkpByItynbVOk5k1Hg.png" /><figcaption>Original response from Agent</figcaption></figure><p>Check the <strong>Responses</strong> field to view full details of generated responses and locate datastore citations. Referencing our previous question, the response is sourced from this location: <strong>“<em>actionLink”: “https://storage.cloud.google.com/shiba-xx/The History of the Shiba Inu Breed in Japan.pdf#page=2”</em></strong></p><pre><br>&quot;Responses&quot;: [<br>              {<br>                &quot;responseType&quot;: &quot;HANDLER_PROMPT&quot;,<br>                &quot;source&quot;: &quot;VIRTUAL_AGENT&quot;,<br>                &quot;payload&quot;: {<br>                  &quot;richContent&quot;: [<br>                    [<br>                      {<br>                        &quot;citations&quot;: [<br>                          {<br>                            &quot;title&quot;: &quot;&quot;,<br>                            &quot;actionLink&quot;: &quot;https://storage.cloud.google.com/shiba-xx/The History of the Shiba Inu Breed in Japan.pdf#page=2&quot;,<br>                            &quot;subtitle&quot;: &quot;Origin of the Name While “inu” means dog in Japanese, “shiba” means brushwood. It&#39;s possible they are named Shiba Inu after the terrain where they hunted or the color of their coat, which is the color of autumn brushwood.&quot;<br>                          },<br>                          {<br>                            &quot;title&quot;: &quot;&quot;,<br>                            &quot;subtitle&quot;: &quot;Shiba means brushwood in Japanese, so it&#39;s possible that the Shiba Inu was named for the terrain where it hunted. It&#39;s also possible that the name came from the Shiba&#39;s coat, which is the same color as the autumn brushwood.&quot;,<br>                            &quot;actionLink&quot;: &quot;https://storage.cloud.google.com/shiba-xx/Shiba inu Dog Breed information.pdf#page=2&quot;<br>                          }<br>                        ],<br>                        &quot;type&quot;: &quot;match_citations&quot;<br>                      }<br>                    ]<br>                  ]<br>                }<br>              },<br>              {<br> # ... (JSON continues) </pre><p>Alright, let’s set up this virtual agent in a Discord channel. After all, a bit of spontaneity never hurts, right?</p><h3>Create a Discord bot</h3><p>I followed the steps on integrating <a href="https://github.com/GoogleCloudPlatform/dialogflow-integrations/tree/master/cx/discord">Dialogflow integrations | Discord</a>. I made changes to the implementation code and you can find it in my repository here: <a href="https://github.com/paulinaMorenoA/dialogflow-integrations/tree/main/cx/discord">Dialogflow CX | Discord integration</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/736/1*A3SOHD0tnDdN-TWScrN88g.png" /><figcaption>Discord ShibaBot</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/746/1*nm6JYW8WP_6wVwXDp0MwWA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/746/1*rhmjT3Sp8CHepnA0oqYQLA.png" /></figure><p>Each response includes a hyperlink to the source, allowing for easy reference to the specific page. The source link’s formatting could be improved, but it did the job. Some answers may be too long and detailed, which is fine for chats but not ideal for Interactive Voice Response (IVR) solutions that need quick and engaging conversations. The question is, how can we make the summary even shorter? This is the ideal time to rely on generators.</p><h3>Adding a Generator</h3><p>We will use a Generator to summarize our response even more. Follow the instructions on <a href="https://cloud.google.com/dialogflow/cx/docs/concept/generative/generators#define">Generators | Dialogflow CX </a>to define a generator. This is what our Generator for this example looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*c6h9lMvKbrD3Sfge" /><figcaption>Generators page</figcaption></figure><p>I’m looking for a deterministic answer so the temperature and top-k values are on the low end but they can be adjusted as needed.</p><p>We need to integrate our Generator to our flow. We need to add a parameter to our Data store that maps to <strong>$request.knowledge.answer[0]</strong> which at run time contains the top answer to the user’s question</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HSrTuCe5znQIA5Zr" /><figcaption>Data stores — Setting parameters</figcaption></figure><p>Remove the default Fulfillment responses</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/781/0*wwBc_cvja1tnJemn" /><figcaption>Data stores — Agent responses</figcaption></figure><p>This parameter becomes a session parameter, meaning you can reference it throughout the current session</p><p>I made a new page named <strong>Summarization Response</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*goKY4ng9oFCRAHmW" /><figcaption>Default Start Flow</figcaption></figure><p>The next step is to incorporate the generator’s response into the entry fulfillment of our new summarization page</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/962/1*axnz3cA7MU9MDUKwngvMAg.png" /><figcaption>Summarization Response page</figcaption></figure><p>We can now refer to <strong>$session.params.response</strong> as the obtained response has been incorporated into the Dialogflow session.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U7ocCjwieW72jQe6RpjrFA.png" /><figcaption>DialogFlow CX- Fulfillment page</figcaption></figure><p>Let’s revisit the question that prompted a detailed response and see if our bot can now summarize it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LZIadELNGYc6uhsRvecXCA.png" /><figcaption>Discord bot</figcaption></figure><p>Thanks to generators, our responses are now short and sweet — just like magic!</p><p>We’ve finished checking out the cool generative AI features of Vertex AI Conversation. If you’re ready to explore creating chatbots and virtual agents that feel more human and less transactional, the time to experiment with Vertex AI Conversation is now. What kind of innovative conversational experiences could you build with this powerful technology?</p><p>Thanks a ton for tuning in! I genuinely hope you found something here that sparked a bit of joy or insight. Keep being awesome, and until next time, happy reading!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=34d8813d3875" width="1" height="1" alt=""><hr><p><a href="https://medium.com/google-cloud/exploring-google-vertex-ai-conversation-dialogflow-cx-generative-ai-features-data-store-34d8813d3875">Exploring Google Vertex AI Conversation — Dialogflow CX Generative AI features Data store &amp;…</a> was originally published in <a href="https://medium.com/google-cloud">Google Cloud - Community</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Guide to Tuning Language Foundation Models in Google Cloud Generative AI Studio]]></title>
            <link>https://medium.com/google-cloud/a-guide-to-tuning-language-foundation-models-in-google-cloud-generative-ai-studio-e47b0d49a43d?source=rss-a6a8466ed041------2</link>
            <guid isPermaLink="false">https://medium.com/p/e47b0d49a43d</guid>
            <category><![CDATA[generative-ai-studio]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[genai-studio-tune-model]]></category>
            <dc:creator><![CDATA[Paulina Moreno]]></dc:creator>
            <pubDate>Wed, 19 Jul 2023 04:50:13 GMT</pubDate>
            <atom:updated>2023-07-20T01:52:25.740Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/713/1*6GkpsfFHSnlA_S8uGdQLxg.png" /></figure><h3>A Guide to Tune Language Foundation Models in Google Cloud Generative AI Studio</h3><p>Generative AI Studio platform gives you access to Google’s large generative AI models, allowing you to test, tune and deploy language foundation models. In this post, we thoroughly explore how to tune a foundation model on the platform, leading to improved performance on specific tasks.</p><p>Whether you’re an experienced practitioner or just starting your journey in Generative AI, this guide will provide you with valuable insights and tips to tuning foundation models.</p><h3>Tune language foundation models on Generative AI Studio</h3><p>Generative AI Studio model tune feature for foundational models is different from the traditional concept of “<strong>fine-tuning</strong>” a model. The goal of <strong>model tuning</strong> is to enhance performance on a <strong>particular task</strong>. When a tuning process is performed, the model gains additional parameters that help encode information to carry out the intended task or learn the desired behavior without <strong>changing the foundational model</strong>.</p><p>During the tuning process, the original model is presented with a <strong>training dataset</strong> that includes various examples of the task. With just a few examples, this training step can significantly improve model performance when tuning for a specific task.</p><h3>Model tuning workflow.</h3><p>At the time of writing this article, the mode tune feature is currently in <strong>public preview</strong> with only <strong>text-bison</strong> model support. For more information please visit <a href="https://medium.com/r?url=https%3A%2F%2Fcloud.google.com%2Fvertex-ai%2Fdocs%2Fgenerative-ai%2Fmodels%2Ftune-models">Generative AI Studio Public Preview Terms</a></p><p>Let’s begin this exciting journey right away, shall we?</p><h3>Step 1- <strong>Prepare your model tuning dataset</strong></h3><p>We need to prepare a training dataset that showcases the task. Our dataset must be in <strong>JSONL</strong> format, where each line contains a single training example. Each example is composed of an <strong>input_text</strong> field and an <strong>output_text</strong> field :</p><pre>{&quot;input_text&quot;: &quot;question: When did the Shiba Inu first arrive in the US? context: Shiba Inu first arrived in the US in the 1950s, but only gained American Kennel Club recognition in 1993. They have become increasingly popular not just in America but around the world for their cat-like personalities and lovable features.&quot;, &quot;output_text&quot;: &quot;The Shiba Inu first arrived in the US in the 1950s.&quot;}</pre><p>One must take into account that the training data aligns with our expected production traffic.</p><p>The same format should be used for the production traffic to maintain consistency in our model’s understanding. In our above example, within the <strong>input_text</strong> field, “<strong>question:”</strong> is followed by <strong>“context:”.</strong> Our production traffic must adhere to the same structure for <strong>input_text</strong>.</p><p>We know we need a few task examples on our training dataset, but what does “few” meant to produce significant results? There is no definitive answer to that question, as it hinges on the task, but it is recommended to provide a minimum of <strong>100 </strong>examples. You can refer to these recommendations for the number of examples — <a href="https://cloud.google.com/vertex-ai/docs/generative-ai/models/tune-models#recommended_configurations">Recommended Configuration for Model Tune</a></p><p>For our working scenario, I opted to engage in a classification task. We will tune the model to classify food products into two categories based on their allergen content, distinguishing between those that contain allergens and those that do not.</p><p>The dataset we are using for the classification task is sourced from Kaggle and consists of information on food ingredients and allergens. In case of your interest, you can view it <a href="https://www.kaggle.com/datasets/uom190346a/food-ingredients-and-allergens">here</a>.</p><p>The <strong>Food Allergens Dataset</strong> contains detailed information about food allergens found in a variety of foods. There are <strong>400</strong> records in the dataset, each corresponding to a food item and listing its associated allergens. I changed the dataset to fit the requirements for the <strong>JSONL</strong> tuning dataset, and it looks like this:</p><pre>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Almond Cookies, Main Ingredient:Almonds, Sweetener:Sugar, Fat[oil]:Butter, Seasoning:Flour&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Chicken Noodle Soup, Main Ingredient:Chicken broth, Sweetener:None, Fat[oil]:None, Seasoning:Salt&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Chicken Noodle Soup, Main Ingredient:Chicken broth, Sweetener:None, Fat[oil]:None, Seasoning:Salt&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Cheddar Cheese, Main Ingredient:Cheese, Sweetener:None, Fat[oil]:None, Seasoning:Salt&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Ranch Dressing, Main Ingredient:Buttermilk, Sweetener:Sugar, Fat[oil]:Vegetable oil, Seasoning:Garlic, herbs&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br>.<br>.<br>.</pre><p>A <strong>Google Cloud Storage</strong> bucket must be created for uploading the training dataset before or during the tuning process.</p><h3><strong>Step 2- Create a Tuned Model</strong></h3><p>We can start the model tuning process through the Google Cloud console, API, or the Vertex AI SDK for Python. Our go-to tool for this example, will be the Google Cloud console. We’ll later cover how to invoke our tuned model from a colab notebook.</p><p>To use <strong>Vertex AI</strong> features in your project for the first time, you must enable the <strong>Vertex AI API</strong>. Then Go to the <strong>Dashboard </strong>on Vertex AI and click the <strong>“Enable all recommended APIs” </strong>button</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UIR8IJFvcky1jJBz1SycqQ.png" /><figcaption>Vertex AI — Dashboard</figcaption></figure><p>Within the <strong>Generative AI Studio</strong> tool, navigate to <strong>Language</strong> and select <strong>“Tune a model.”</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Se2mDF5UqHBNEghD0ao_RA.png" /><figcaption>Vertex AI — Generative AI Studio / Language</figcaption></figure><p>You can either upload the <strong>JSONL</strong> tuning dataset to an existing bucket or create a new one by selecting the <strong>“Upload JSONL file to Cloud Storage”</strong> option.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E1rEoo9xb_3Lv7Ovq54hSQ.png" /><figcaption>Tune Model — Tuning dataset</figcaption></figure><p>In the <strong>Model details</strong> section, you can name your tuned model, and keep in mind that <strong>“text-bison”</strong> is the only option for the base model(for now). Train steps <a href="https://cloud.google.com/vertex-ai/docs/generative-ai/models/tune-models#recommended_configurations">recommendations</a> for classification task are 100–500 steps.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oDMskov6ZY4wr_qVSVPMtA.png" /><figcaption>Tune Model — Model details</figcaption></figure><p>For the training steps I decided to go with 400 steps and leave the “Learning Rate” field with the default value.</p><p>Once you click <strong>Start Tuning</strong>, a <strong>pipeline job would be created.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_EX1ywCfUrdECnWie2llWw.png" /><figcaption>Tuning Model — List of tuned models</figcaption></figure><p><strong>To</strong> check the status of your model tuning job, click the link under the <strong>Pipeline run </strong>column, this link will take you to <strong>Vertex AI Pipelines</strong> page. The pipeline run page appears and displays the pipeline’s runtime graph.</p><p>For now tuning can only happen in the <strong>“europe-west4”</strong> region and the new model can only be deployed in the <strong>“us-central1”</strong> region</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*841E4GBymKEH0_g3-QXvNQ.png" /><figcaption>Vertex AI Pipelines</figcaption></figure><p>The pipeline summary shows the basic information for pipeline steps, this information includes <strong>execution details</strong>, the <strong>input parameters</strong> that were passed to the step, and any <strong>output parameters</strong> that the step passed to the pipeline.</p><p>Our model tuning process requires thorough attention and careful execution so it will take some time to finish. You can decide to continuously monitor or Alternatively, you can <a href="https://cloud.google.com/vertex-ai/docs/pipelines/email-notifications">configure email notifications</a> for <strong>Vertex AI Pipelines</strong> so you are notified by email when the model tuning job finishes or fails.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LV5U2NYpR-mg7h5-XSr8mg.png" /><figcaption>Vertex AI Pipeline graph</figcaption></figure><p>To view your tuned models in the Google Cloud console, go to the <strong>Vertex AI Model Registry</strong> page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*53YghxZPvtUIAirMA8A91w.png" /><figcaption>Vertex AI Model Registry</figcaption></figure><h3>Step 3 — <strong>Use your Tuned Model</strong></h3><p>The tuned model is ready, and now it’s time to use it. Navigate to the <strong>Language</strong> section of <strong>Generative AI Studio</strong> where you can craft a new text prompt. Our model will be available to you to choose under the <strong>Model</strong> dropdown selection. We will explore the option of calling our model with the <strong>Python Vertex AI SDK</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5oglJZ4g7oMWg4WJt2I5bg.png" /><figcaption>Language text prompt UI</figcaption></figure><p>If you’re unfamiliar with <strong>text prompts</strong>, <a href="https://cloud.google.com/vertex-ai/docs/generative-ai/learn/generative-ai-studio">Generative AI Studio</a> documentation is an excellent resource to learn more about this remarkable feature.</p><p>Choosing our model is just the beginning. There are other configuration parameters to consider. <strong>Click the question mark</strong> for a comprehensive explanation of each parameter, and if that’s not enough, the <a href="https://cloud.google.com/vertex-ai/docs/generative-ai/learn/introduction-prompt-design#experiment-with-different-parameter-values">Introduction to prompt design documentation</a> will offer you additional insights.</p><p>Let’s try our model with some examples.</p><p>Roughly <strong>20%</strong> of the total data from the Food Allergens dataset was separated into a <strong>“testing”</strong> subset. We will use the following examples from those records for our next text prompts:</p><pre>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Chocolate Chip Pancakes, Main Ingredient:Flour, Sweetener:Sugar, Fat[oil]:Butter, Seasoning:Chocolate chips&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br><br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Chicken Biryani, Main Ingredient:Chicken, Sweetener:None, Fat[oil]:Ghee, Seasoning:Basmati rice, spices&quot;, &quot;output_text&quot;: &quot;Does not contain&quot;}<br><br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Hawaiian Pizza, Main Ingredient:Pizza dough, Sweetener:None, Fat[oil]:None, Seasoning:Pineapple, ham&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br><br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Chocolate Chip Pancakes, Main Ingredient:Flour, Sweetener:Sugar, Fat[oil]:Butter, Seasoning:Chocolate chips&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br><br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:Beef Wellington, Main Ingredient:Beef, Sweetener:None, Fat[oil]:Butter, Seasoning:Mushrooms, puff pastry&quot;, &quot;output_text&quot;: &quot;Contains&quot;}<br><br>{&quot;input_text&quot;: &quot;Given the following Food Product information classify it into one of the following classes: [Contains, Does not contain] allergens Food Product:&quot;, &quot;output_text&quot;: &quot;Does not contain&quot;}</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zaLkZ5gwM05ntSNyDQroSg.png" /><figcaption>Example 1</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-OLnnXoousRRRHIy7a22cg.png" /><figcaption>Example 2</figcaption></figure><p>As our model is now optimized to detect allergens in food products, we refrained from passing any examples from the UI and we’ve lowered the <strong>Temperature</strong> parameter. The reason behind this is that <strong>lower temperatures</strong> are best suited for <strong>classification</strong> tasks requiring concrete, non-creative responses. To produce the most accurate results, <strong>low temperature and low top-K </strong>values are recommended because of the task’s deterministic nature.</p><p><strong>Take into account:</strong> The way you construct your prompt can influence the model towards producing the intended result (or not)</p><p>We can enhance our text prompt by providing a list of food allergens instead of individual ones.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zpfVdQ7pY4ucB-ybLafRlw.png" /><figcaption>Example 3</figcaption></figure><p>The output from our model matches the expected results based on our previous test examples. Our next step is to recreate these examples using the <strong>Python Vertex AI SDK</strong>.</p><p>From your text prompt you have a <strong>“&lt;&gt;VIEW CODE”</strong> option on the top right side. After selecting the view code option, select <strong>“PYTHON COLAB”</strong>, and you will be shown a sample Python code to use with a colaboratory notebook.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iZ4JlEyklnLw4Ar28UlNtA.png" /><figcaption>Python Colab sample code</figcaption></figure><p>Using the sample code as a starting point, I added some additional lines and created a colab notebook. To illustrate its functionality, I included the last four records we tested before.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-rJ8UKtCwj-sErYo_-Vb0A.png" /><figcaption>Tune Model — colab notebook</figcaption></figure><p>Below is the full code from the previous notebook.</p><pre>!pip install google-cloud-aiplatform<br>!pip install &quot;shapely&lt;2.0.0&quot;<br><br>import IPython<br><br>app = IPython.Application.instance()<br>app.kernel.do_shutdown(True)<br><br>from google.colab import auth as google_auth<br>google_auth.authenticate_user()<br><br>from google.colab import auth as google_auth<br>google_auth.authenticate_user()<br><br>import vertexai<br>from vertexai.preview.language_models import TextGenerationModel<br><br>vertexai.init(project=&quot;1055236240165&quot;, location=&quot;us-central1&quot;)<br>parameters = {<br>    &quot;temperature&quot;: 0.1,<br>    &quot;max_output_tokens&quot;: 256,<br>    &quot;top_p&quot;: 0.8,<br>    &quot;top_k&quot;: 3<br>}<br>model = TextGenerationModel.from_pretrained(&quot;text-bison@001&quot;)<br>model = model.get_tuned_model(&quot;projects/1055236240165/locations/us-central1/models/546019673376817152&quot;)<br>response = model.predict(<br>      &quot;&quot;&quot;Given the following list of food products information classify each one of them into one of the following classes: [Contains, Does not contain] allergens .Give results in JSON with Food product and Allergens values.<br><br>list [<br>input: Food Product:Hawaiian Pizza, Main Ingredient:Pizza dough, Sweetener:None, Fat[oil]:None, Seasoning:Pineapple, ham<br>input: Food Product:Chocolate Chip Pancakes, Main Ingredient:Flour, Sweetener:Sugar, Fat[oil]:Butter, Seasoning:Chocolate chips<br>input: Food Product:Beef Wellington, Main Ingredient:Beef, Sweetener:None, Fat[oil]:Butter, Seasoning:Mushrooms, puff pastry<br>input: Food Product:Greek Lemon Potatoes, Main Ingredient:Potatoes, Sweetener:None, Fat[oil]:Olive oil, Seasoning:Lemon juice, herbs<br>]<br>&quot;&quot;&quot;,<br>    **parameters<br>)<br>print(f&quot;Response from Model: {response.text}&quot;)</pre><p>Summing up, with <a href="https://cloud.google.com/vertex-ai/docs/generative-ai/learn/overview"><strong>Google Cloud Generative AI Studio</strong></a><strong> </strong>model tuning, even those with limited AI expertise can easily improve the performance of a foundation model for particular tasks in just a few steps. Being a part of the journey during the product’s public preview is an exciting time, and providing feedback can make the experience even more rewarding.</p><p>Thank you for reading and happy model tuning!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e47b0d49a43d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/google-cloud/a-guide-to-tuning-language-foundation-models-in-google-cloud-generative-ai-studio-e47b0d49a43d">A Guide to Tuning Language Foundation Models in Google Cloud Generative AI Studio</a> was originally published in <a href="https://medium.com/google-cloud">Google Cloud - Community</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>