<?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 Anandan on Medium]]></title>
        <description><![CDATA[Stories by Anandan on Medium]]></description>
        <link>https://medium.com/@skabhianan?source=rss-2b89a0dd3bf9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*vfexSmvp72Mx4eUiFwQTwQ.jpeg</url>
            <title>Stories by Anandan on Medium</title>
            <link>https://medium.com/@skabhianan?source=rss-2b89a0dd3bf9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 00:58:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@skabhianan/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[NotebookLM: The Grounded AI Research and Synthesis Partner]]></title>
            <link>https://medium.com/@skabhianan/notebooklm-the-grounded-ai-research-and-synthesis-partner-4117e0f7c2f2?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/4117e0f7c2f2</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Tue, 16 Jun 2026 14:13:30 GMT</pubDate>
            <atom:updated>2026-06-16T14:13:30.588Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s a fantastic learning experience in learning about Google’s NotebookLM by uploading the Google’s official website link itself , a few youtube video tutorials, and interestingly adding the NotebookLM’s output as a source using the “convert to source” feature in the “studio” section to the NotebookLM. It has beautifully summarized!! There are automated and suggested possible questions. There are a few questions asked from myend. Here we go!!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hBYU07psW4s2t_z18fSjAw.png" /></figure><blockquote>The grounded relevant documents:</blockquote><ol><li><a href="https://notebooklm.google/">https://notebooklm.google/</a></li><li><a href="https://youtu.be/-Nl6hz2nYFA">NotebookLM Will Change How You Learn — Here’s Why!</a></li><li><a href="https://youtu.be/FOs4RDTC52Q">The Ultimate Guide to NotebookLM — All 2025 Features Explained</a></li></ol><blockquote>The chat with NotebookLM on NotebookLM and it’s response based on the grounded documents:</blockquote><p><em>Q: Provide executive summary for NotebookLM from the Google’s official link</em></p><p>A: According to Google’s official overview, NotebookLM is an <strong>AI research tool and thinking partner</strong> designed to help users understand complex information by summarizing sources and identifying connections between topics.</p><p>It functions by allowing users to upload specific documents — including PDFs, websites, YouTube videos, audio files, and Google Docs or Slides — and then querying that specific knowledge base.</p><p><strong>Core Functionality and Outputs:</strong></p><p>NotebookLM transforms source material into various digestible formats to reinforce understanding, such as:</p><ul><li><strong>Audio Overviews:</strong> Interactive, podcast-style audio discussions that summarize provided materials.</li><li><strong>Customized Reports:</strong> Tailored documents that synthesize insights from the sources.</li><li><strong>Study and Planning Aids:</strong> The tool can generate flashcards, quizzes, and dynamic reports to help users prepare for meetings or exams.</li><li><strong>Creative Support:</strong> It assists in organizing research and brainstorming or drafting scripts for projects like podcasts.</li></ul><p><strong>Key Advantages:</strong></p><ul><li><strong>Source Grounding:</strong> As a research assistant, its primary advantage is being “source-grounded,” meaning responses are strictly based on the information provided by the user. This transparency builds trust and <strong>significantly reduces the likelihood of AI hallucinations</strong> or errors.</li><li><strong>Nuanced Understanding:</strong> Built with the latest Gemini models, the tool can identify deep connections across multiple documents and explain complex concepts in simple terms.</li><li><strong>Efficiency:</strong> It dramatically speeds up the process of synthesizing and extracting insights from large volumes of text, helping users move past the stage of “searching for the answer” to active drafting and application.</li></ul><p><strong>Privacy and Security</strong></p><p>Google emphasizes that <strong>user privacy is a priority</strong>. For organizations and schools, data remains private and is not used to train Google’s models. For individual users, data is not used for training unless the user explicitly chooses to share feedback. Furthermore, NotebookLM does not share data with third parties.</p><p><strong>Practical Applications</strong></p><p>NotebookLM is designed to integrate into daily productivity workflows by:</p><ul><li>Reviewing meeting notes and reports to prepare for the day.</li><li>Acting as a research assistant to clarify concepts or support the drafting of emails and proposals.</li><li>Summarizing daily notes to generate to-do lists and recaps</li></ul><p><em>Q: Navigating the NotebookLM Interface and Functional Architecture</em></p><p>A: NotebookLM is organized into a primary three-panel interface designed for research, synthesis, and content creation. The following are the key components and their specific usages:</p><p><strong>The Three-Panel Interface</strong></p><ul><li><strong>Sources Panel (Left):</strong> This panel is used to <strong>upload and manage the data</strong> the AI uses to ground its responses. You can add PDFs, website and YouTube URLs, Google Docs, Slides, and audio files. Clicking a source opens a <strong>Source Guide</strong>, which provides a one-paragraph summary of that specific material.</li><li><strong>Chat Panel (Middle):</strong> This is the main area for <strong>interacting with the AI</strong>. Users can ask specific questions about their sources, and the AI provides bulleted or detailed summaries. A critical feature here is <strong>inline citations</strong> — numbered links that, when clicked, take you directly to the original source and highlight the specific passage referenced.</li><li><strong>Studio Panel (Right):</strong> This functions as a <strong>creative workspace</strong> for saving insights and generating high-level outputs. It is used to:</li><li><strong>Save Notes:</strong> Keep valuable AI responses as “Evergreen insights” for long-term use.</li><li><strong>Create Structured Documents:</strong> Generate briefing documents, FAQs, timelines, and study guides with a single click.</li></ul><p><strong>Core Functional Components</strong></p><ul><li><strong>Audio Overviews:</strong> This tool generates an <strong>interactive, podcast-style conversation</strong> between two AI hosts who summarize and discuss your sources.</li><li><strong>Interactive Mode:</strong> Allows you to “join” the podcast in real-time to ask the hosts questions or redirect their focus.</li><li><strong>Video Overviews:</strong> A brand-new feature that creates a <strong>narrated slideshow video</strong>, transforming your sources into a visual presentation with voiceover.</li><li><strong>Mind Map:</strong> Used to create a <strong>visual breakdown of concepts</strong>. It allows you to see connections between ideas step-by-step and can be downloaded as a PNG file for external use.</li><li><strong>Discover Sources:</strong> A search component used to <strong>find additional research material</strong> from across the web, such as Wikipedia or academic articles, which can then be imported directly into the notebook.</li></ul><p><strong>Notebook Settings &amp; Configuration:</strong></p><ul><li><strong>Output Language:</strong> Allows you to change the language in which the AI communicates.</li><li><strong>Configure Notebook:</strong> Used to define the <strong>conversational style</strong> (e.g., creative vs. analytical) and the preferred length of AI responses.</li><li><strong>Convert to Source:</strong> This feature allows you to turn a generated note or report into a <strong>primary source</strong>, enabling the AI to cite and reference its own summarized content in future queries.</li></ul><p><strong>Advanced and Plus Components</strong></p><ul><li><strong>Feature Notebooks:</strong> Public notebooks curated by experts that users can explore to see how sources are organized and queried.</li><li><strong>Chat Modes (Plus Version):</strong> Used to assign the AI a <strong>specific persona</strong>, such as a “Marketing Expert,” “Analyst,” or “Life Coach,” to tailor the tone and depth of its insights.</li><li><strong>Collaboration Tools (Plus Version):</strong> Allows users to <strong>share entire notebooks</strong> (including all sources and notes) or specific “chat-only” access with team members.</li></ul><blockquote><strong>Architecting the NotebookLM Knowledge Base</strong></blockquote><p><em>Q: List out the possible sources that can be added to the NotebookLM</em></p><p>A: You can add a wide variety of information to NotebookLM to create a personalized knowledge base. The possible sources include:</p><ul><li><strong>Document Files:</strong> You can upload <strong>PDFs</strong> directly from your computer.</li><li><strong>Google Drive Content:</strong> You can import <strong>Google Docs</strong> and <strong>Google Slides</strong>.</li><li><strong>Web-Based Material:</strong> You can add <strong>website URLs</strong> and <strong>YouTube video URLs</strong>. The tool will analyze the text from the websites or the transcripts from the videos.</li><li><strong>Audio Files:</strong> The platform supports <strong>audio files</strong>, including personal <strong>voice memos</strong>.</li><li><strong>Manual Text Entry:</strong> You can use the <strong>“copied text”</strong> feature to paste any text directly into a source window, such as notes, journal entries, or snippets from other documents.</li><li><strong>Visual and Complex Data:</strong> Because the tool is multimodal, it can process <strong>images, charts, tables, and diagrams</strong> that are embedded within <strong>Google Slides</strong>.</li><li><strong>Discovered Sources:</strong> You can use the <strong>“Discover sources”</strong> tool to search for and import new articles and PDFs from the web directly into your notebook.</li></ul><p>In terms of capacity, you can add up to <strong>50 individual sources</strong> to a single notebook. Each of these sources can contain up to <strong>500,000 words</strong>, allowing for a total context window of <strong>25 million words</strong> in the standard version and <strong>50 million words</strong> in the Plus version</p><blockquote><a href="https://www.youtube.com/watch?v=-Nl6hz2nYFA&amp;t=39s"><strong>NotebookLM Will Change How You Learn — Here’s Why!</strong></a></blockquote><p><em>Q: List out the 5 insane features the author is discussing in this video walk-through!!</em></p><p>A: In the video walk-through, the author highlights the following five major features of NotebookLM:</p><ol><li><strong>Audio Overviews:</strong> This feature allows you to generate a <strong>podcast-style conversation</strong> between two AI hosts who discuss your sources. It includes an <strong>interactive mode</strong> where you can join the conversation in real-time to ask questions or provide additional context.</li><li><strong>Expanded Context Window:</strong> Using the Gemini 2.0 Flash model, NotebookLM now has a massive “short-term memory” capable of processing up to <strong>25 million words</strong> across 50 different sources. This allows the AI to analyze huge repositories of information, such as years of personal notes or thousands of customer interviews, without losing the thread of the data.</li><li><strong>Multimodal Sources:</strong> NotebookLM can now process a wide variety of formats beyond text, including <strong>YouTube videos, audio files, and Google Slides</strong>. It can even analyze <strong>complex visual data</strong> like charts, tables, and diagrams within slide decks.</li><li><strong>Redesigned Interface:</strong> The updated layout is organized into three panels: <strong>Sources</strong> (left), <strong>Chat</strong> (middle), and the <strong>Studio panel</strong> (right) for notes and audio overviews. A key part of this redesign is the inclusion of <strong>inline citations</strong>, which link the AI’s responses directly to the original source material to prevent hallucinations.</li><li><strong>Audio Overviews:</strong> This feature allows you to generate a <strong>podcast-style conversation</strong> between two AI hosts who discuss your sources. It includes an <strong>interactive mode</strong> where you can join the conversation in real-time to ask questions or provide additional context.</li></ol><blockquote><strong>The Architecture of Source Grounding</strong></blockquote><p><em>Q: How does notebook source grounding eliminate AI hallucinations?</em></p><p>A: NotebookLM addresses the problem of AI hallucinations through a philosophy called <strong>source grounding</strong>, which ensures the AI’s responses are based specifically on the material you provide. Here is how this process works to eliminate errors and build trust:</p><ul><li><strong>Focus on Short-Term Memory:</strong> While traditional AI models rely on “long-term memory” (their broad training data from the internet), NotebookLM prioritizes the <strong>context window</strong>, or “short-term memory”. This context window contains the specific documents, websites, and notes you have uploaded as sources. By restricting the AI to this specific set of data, it is much less likely to make things up or guess.</li><li><strong>Inline Citations:</strong> Every statement the AI makes in the chat is accompanied by <strong>small numbers representing source citations</strong>. This provides immediate transparency regarding where the information originated.</li><li><strong>Direct Source Verification:</strong> You can <strong>double-check the AI’s accuracy</strong> by clicking on these citation numbers. Doing so automatically opens the original source — whether it is a PDF, a website, or a YouTube transcript — and <strong>highlights the exact passage</strong> or data point the AI is referring to.</li><li><strong>Multimodal Grounding:</strong> This grounding also applies to complex visual data. If you upload a slide deck with intricate charts or tables, the AI can analyze that specific image and provide a citation that takes you directly to the <strong>specific table or diagram</strong> used to generate the answer.</li></ul><p>Because every claim is anchored to a specific piece of evidence in your sources, the AI effectively “not hallucinate or very, very rarely” does so, allowing you to treat it as a reliable tool for high-stakes decision-making</p><blockquote><strong>The Architecture of Artificial Intelligence Memory</strong></blockquote><p><em>Q: What is the difference between long-term and short-term AI memory?</em></p><p>A: Based on the sources, AI memory is divided into two distinct categories that function similarly to human memory:</p><ul><li><strong>Long-Term Memory:</strong> This refers to the AI’s <strong>training data</strong>, which consists of all the information used during its initial creation. Currently, most major AI models share very similar long-term memory because they are all trained on roughly the same data: the entire internet. Because this information is the same across most platforms, it provides no real unique advantage to the user.</li><li><strong>Short-Term Memory:</strong> This is the specific data the AI is working with in the present moment, such as the current conversation, the specific person it is talking to, or the particular problem it is trying to solve. In NotebookLM, this is referred to as the <strong>context window</strong>. This memory is supplied by you in the form of documents, notes, and sources.</li></ul><p>The primary advantage of modern AI tools like NotebookLM comes from the <strong>short-term memory</strong> because it contains the specifics of your project — such as your company’s history, your personal notes, or unique skills — which the AI uses to provide personalized answers rather than generic guesses. As previously discussed, this reliance on short-term memory is what enables <strong>source grounding</strong>, ensuring the AI remains accurate and avoids hallucinations by anchoring its responses to the specific material you have provided.</p><blockquote><strong>The Architecture of Contextual Grounding</strong></blockquote><p><em>Q: What is the role of the context window in grounding?</em></p><p>A: The <strong>context window</strong> serves as the AI’s <strong>“short-term memory,”</strong> and it is the foundational space where <strong>source grounding</strong> occurs to ensure accuracy.</p><p>The role of the context window in grounding can be understood through several key functions:</p><ul><li><strong>Replacing “Guessing” with Specific Data:</strong> Unlike traditional AI that relies on “long-term memory” (general training data from the internet), grounding forces the AI to operate within its “short-term memory” — the specific sources you have provided. This allows the AI to provide recommendations and answers based on <strong>your specific history, documents, and notes</strong> rather than generic patterns.</li><li><strong>Eliminating Hallucinations:</strong> Because the AI is restricted to the information contained within this context window, it is “basically <strong>not going to hallucinate</strong> or very, very rarely”. It does not need to fill in gaps with made-up information because it has the exact data it needs right in front of it.</li><li><strong>Enabling Massive Repositories for Analysis:</strong> Modern context windows in NotebookLM can take in up to <strong>25 million words</strong> (standard) or <strong>50 million words</strong> (Plus version) across dozens of sources. This massive window allows you to “point the AI at a massive repository of information” and let it identify recurring patterns, connections, and insights across your entire personal or professional history.</li><li><strong>Facilitating Verifiable Citations:</strong> The grounding within the context window is what enables <strong>inline citations</strong>. Because the AI’s response is anchored to specific parts of the “short-term memory,” users can click a citation number and be taken directly to the <strong>highlighted passage in the original source</strong> to verify the information themselves.</li><li><strong>Acting as a “World-Class Teacher”:</strong> By having a large context window, the AI acts as a teacher with a “fantastic memory” that can follow the thread of a topic, reference previous parts of a conversation, and plan personalized learning based on everything you have read and consumed.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4117e0f7c2f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Conversational Data Agents for Cloud SQL for PostgreSQL]]></title>
            <link>https://medium.com/@skabhianan/conversational-data-agents-for-cloud-sql-for-postgresql-c770ef90cd5c?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c770ef90cd5c</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Mon, 15 Jun 2026 17:11:52 GMT</pubDate>
            <atom:updated>2026-06-15T17:11:52.286Z</atom:updated>
            <content:encoded><![CDATA[<h4>Unlocking Your Database: A Guide to Data Agents for Cloud SQL for PostgreSQL</h4><p><strong>Note:</strong> This feature is currently in <strong>Preview</strong> and is subject to “Pre-GA Offerings Terms”</p><blockquote>What is a Conversational Data Agent?</blockquote><p>Conversational data agents let you interact with your database data using a natural language interface. By building these agents, you help users “talk to their data,” unlocking insights from operational databases without requiring them to write complex SQL queries.</p><p>At a high level, a data agent is a combination of persona, a set of data sources, with access to a set of business knowledge critical for its purposes.</p><p>For application developers, these agents provide the following benefits:</p><ul><li>High accuracy: by using authored context, you can achieve high accuracy for specific business questions. <em>Authored context</em> is the primary key factor that agent creators can use to improve accuracy. It includes schema descriptions, system instructions, and structured context that provides additional information about expected database queries.</li><li>Reduced complexity: agents translate natural language into SQL queries, execution, and even data summarization or visualization.</li><li>Flexibility: you can draft agents for personal testing, or you can publish them to make them available to other users in your project or programmatically using the API.</li></ul><blockquote>Conversational analytics for Cloud SQL for PostgreSQL overview</blockquote><p>Conversational analytics in Cloud SQL for PostgreSQL lets you chat with agents about your database data using natural language. To get answers about your data, you:</p><ul><li>Create <em>data agents</em> for a set of knowledge sources, such as tables and views, that you select.</li><li>[Optional] Create context and instructions for an agent to configure the data agent to effectively answer questions for specific use cases.</li></ul><blockquote>Guided vs. Advanced Development Methods</blockquote><p>Google Cloud offers two pathways for building these agents:</p><ul><li><strong>Guided Development:</strong> This method is performed directly within the <strong>Google Cloud Console</strong>. It is ideal for users who prefer a no-code experience and want to get an agent operational quickly using automated context generation.</li><li><strong>Advanced Development:</strong> This approach uses the <strong>Gemini CLI</strong> or an IDE. It is recommended for developers who need deeper control over agent quality and want to manually refine the semantic translation layer (Context Sets)</li></ul><p>In this blog, we will walk-through the Guided Development from the Google Cloud Console.</p><blockquote>The core components:</blockquote><h3>Data agents</h3><p>Data agents consist of one or more knowledge sources, and a set of instructions specific to a use case for processing that database data. When you create a data agent, you can configure it using the following options:</p><ul><li>Use <em>knowledge sources</em> such as tables and views with a data agent.</li><li>Provide custom table and field metadata to describe the database data in the most appropriate way for the given use case.</li><li>Provide instructions for interpreting and querying the data, such as defining the following:</li><li>Synonyms and business terms for field names</li><li>Most important fields and defaults for filtering and grouping</li><li>Using either guided or advanced context generation, provide structured contexts that the data agent can use to shape an agent’s response structure and to learn the business logic that your organization uses.</li></ul><h3>Manage data agents</h3><p>You can create, manage, and work with the following types of data agents in the <strong>Agent</strong> tab in the Google Cloud console:</p><ul><li>A predefined sample agent for each Google Cloud project.</li><li>A list of your drafted, created, and published agents.</li><li>A list of agents that other people create and share with you.</li></ul><p>For more information, see <a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/sql-postgres/create-data-agents">Create data agents</a>.</p><p>Other services in the project that support data agents, such as the <a href="https://docs.cloud.google.com/gemini/docs/conversational-analytics-api/overview">Conversational Analytics API</a>, can access data agents that you create in Cloud SQL for PostgreSQL . You can also access an agent created in the Google Cloud console by calling it using the Conversational Analytics API.</p><h3>Conversations</h3><p>Conversations are persisted chats with a data agent or database data source. You can ask data agents multi-part questions that use common terms like “sales” or “most popular,” without having to specify table field names or define conditions to filter the database data.</p><p>The chat response returned to you provides the following features:</p><ul><li>The answer to your question as text, code, or charts (where appropriate)</li><li>The agent’s reasoning behind the results.</li><li>Metadata about the conversation, such as the agent and database data sources used.</li></ul><p>When you create a direct conversation with a database data source, the <a href="https://docs.cloud.google.com/gemini/docs/conversational-analytics-api/overview">Conversational Analytics API</a> interprets your question without the context and processing instructions that a data agent offers. Because of this, direct conversation results can be less accurate. Use data agents for cases that require greater accuracy.</p><p>You can create and manage conversations in Cloud SQL for PostgreSQL using the Google Cloud console. For more information, see <a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/sql-postgres/create-conversations">Analyze data with conversations</a>.</p><p><strong>Step 1: Navigating to Data Agents (preview)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4MJalFZCKz3eJ720rO0VLQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OxeoUjx8aQ9A5mIiKNurYg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cZJNZMIWaMSKIFBv0W37YA.png" /></figure><p><strong>Step 2: Initiating Data Agent Creation</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nHjkkAMbdMKeITQqv_k9NA.png" /></figure><p><strong>Step 3: Configuring Agent Basics</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_BYS_hm1ulJmRFQzJfq-HA.png" /></figure><p><strong>Step 4: Selecting Knowledge Sources</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mYX631xpuuh03d6iMmoY9w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hRE1ywULErnarEDCTjCjag.png" /></figure><p><strong>Step 5: Finalizing and Publishing the Agent</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TYGikJLZ9o1cJavDpKGVRQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OZwaIvZEzvTDMB_kg3IiVA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_dk1pkpY6gwA0n1O66UFXQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MErZe1dZGgTuoPShF0MF0Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fj5B1L8KvSX1u3dbysBJZg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*MErZe1dZGgTuoPShF0MF0Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4zcdPfccOuwKcfU4fO6j1w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*wF_oOd47tbh6Kvac2iDvYA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RvCrDVfoUq9dBgj3ICED8w.png" /></figure><blockquote>References:</blockquote><p><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/sql-postgres">Conversational analytics for Cloud SQL for PostgreSQL overview</a></p><p><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics-api/overview#setup">Conversational Analytics API: Build data agents and chat with your data</a></p><p><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/sql-postgres/create-data-agents">Create data agents</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c770ef90cd5c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Conversational analytics with Google Products]]></title>
            <link>https://medium.com/@skabhianan/conversational-analytics-with-google-products-02751397ba97?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/02751397ba97</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Sat, 13 Jun 2026 15:48:51 GMT</pubDate>
            <atom:updated>2026-06-13T15:48:51.836Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Conversational analytics overview</blockquote><p>Conversational analytics is a chat-with-your-data feature that you can use to do the following:</p><ul><li>Create simple data agents to chat about your data using natural language.</li><li>Create data agents that include context and instructions to effectively answer questions for specific use cases.</li><li>Create data agents that you make available to users in your organization.</li><li>Call the underlying <a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics-api/overview">Conversational Analytics API</a> directly to create customized data agents or agentic applications that integrate data from different products or address complex use cases.</li></ul><p>Conversational analytics is available in the following Google Cloud products:</p><ul><li><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/alloydb">AlloyDB for PostgreSQL</a></li><li><a href="https://docs.cloud.google.com/bigquery/docs/conversational-analytics">BigQuery</a></li><li><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/sql-mysql">Cloud SQL for MySQL</a></li><li><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/sql-postgres">Cloud SQL for PostgreSQL</a></li><li><a href="https://docs.cloud.google.com/looker/docs/conversational-analytics-overview">Looker</a></li><li><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics/spanner">Spanner</a></li></ul><blockquote>Conversational Analytics API architecture and key concepts</blockquote><p>Developers can use the Conversational Analytics API, which is accessed through geminidataanalytics.googleapis.com, to build an artificial intelligence (AI)-powered chat interface, or <em>data agent</em>.</p><p>The API uses natural language to answer questions about structured data in BigQuery, Looker, and Data Studio, and also supports querying data from AlloyDB, GoogleSQL for Spanner, Cloud SQL, and Cloud SQL for PostgreSQL through the new QueryData method.</p><p>With the Conversational Analytics API, you provide your data agent with business information and data (<em>context</em>), as well as access to tools such as SQL, Python, and visualization libraries. These agent responses are presented to the user and can be logged by the client application, creating a seamless and auditable data chat experience.</p><p><em>The following diagram illustrates the stages of an agent’s workflow when a user asks a question:</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*dUdRW5GSBK0uC98p.png" /></figure><p>As shown in the diagram, when a user asks a question, the agent processes the request in the following stages:</p><ol><li><strong>Input</strong>: The user submits a question in natural language, along with any additional context that you provided previously.</li><li><strong>Data sources</strong>: The agent connects to your data in Looker, BigQuery, and Data Studio for chat capabilities. It can also query data from AlloyDB, GoogleSQL for Spanner, Cloud SQL for MySQL, and Cloud SQL for PostgreSQL databases by using the QueryData method.</li><li><strong>Reasoning engine</strong>: The core of the agent processes the user’s question by using available tools to generate an answer.</li><li><strong>Output</strong>: The agent returns a stream of messages, which can contain text, data, or charts. For some data sources, text messages provide step-by-step insight into the agent’s reasoning, report progress on an action, or provide the final answer to you.</li></ol><blockquote><strong>The involved components &amp; detailed interrelations of them:</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/629/1*hs2lE-hZMlroTctlsDZaWQ.png" /></figure><p>1. Conversational Analytics (The Interface &amp; Dialogue Engine)</p><ul><li><strong>What it does:</strong> This is the entry point for end-users. It leverages technologies like Natural Language Processing (NLP) to manage the state of user sessions, resolve ambiguities, and clarify user intents.</li><li><strong>Relation to others:</strong> When a user submits a prompt, Conversational Analytics translates it into an actionable business goal and passes the conversational context forward to the <strong>Data Agent</strong>.</li></ul><p>2. Data Agent (The Orchestrator / LLM Controller)</p><ul><li><strong>What it does:</strong> The Data Agent is an intelligent autonomous coordinator powered by a Large Language Model (LLM). Its primary job is to bridge the massive gap between human communication and relational databases.</li><li><strong>Relation to others:</strong> The agent acts as the brain. It evaluates incoming requests from the <strong>Conversational Analytics</strong> layer, checks against the constraints defined within the <strong>Context Sets</strong> to generate correct database syntax, and packages the final payload into a structured format represented as <strong>QueryData</strong>.</li></ul><p>3. Context Sets (The Knowledge Base &amp; Governance Layer)</p><ul><li><strong>What it does:</strong> Context Sets act as the ground-truth dictionary for the AI. Databases are notoriously abstract (e.g., a column titled rev_q3_f doesn’t mean anything inherently to a base LLM). Context Sets map schema definitions, data types, primary/foreign key relationships, and row-level security parameters.</li><li><strong>Relation to others:</strong> The <strong>Data Agent</strong> dynamically injects these Context Sets into its prompt window during a Retrieval-Augmented Generation (RAG) loop. This prevents hallucinations and ensures the generated code matches the exact syntax variations specific to either <strong>Cloud SQL for MySQL</strong> or <strong>Cloud SQL for PostgreSQL</strong>.</li></ul><p>4. QueryData (The Data Payload &amp; Execution Instance)</p><ul><li><strong>What it does:</strong> QueryData represents the live transactional payload state. It encompasses the dynamically compiled query string (e.g., a specific dialect of PostgreSQL or MySQL dialect optimization) along with the operational metadata and the raw dataset records fetched from the database storage layer.</li><li><strong>Relation to others:</strong> The <strong>Data Agent</strong> pushes the compiled SQL query block of the QueryData object to the Cloud SQL database instance. Once Cloud SQL evaluates the request, the resulting row matrix is stuffed back into the QueryData object. The <strong>Data Agent</strong> inspects this data and hands it back to the <strong>Conversational Analytics</strong> engine to format the final visual response (like chart generation or tabular reporting) for the human user.</li></ul><blockquote>Workflows for designing and using agents</blockquote><p>The Conversational Analytics API supports workflows for <em>agent creators</em> (who build and configure agents) and for <em>agent consumers</em> (who interact with existing agents).</p><p>The following diagram illustrates the end-to-end process, from the initial setup by an agent creator to the final interactions with an agent consumer:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*CgegKwXPe7j3Gc7c.png" /></figure><blockquote>Conversation modes</blockquote><p>Conversational Analytics API agents support different conversation modes that determine how an agent handles conversation history and the persistence of context across interactions. The following conversation modes are available:</p><ul><li><strong>Stateless mode</strong>: The agent doesn’t store conversation history. Each interaction is treated independently. This mode is useful for applications where you don’t need to maintain context across multiple turns.</li><li><strong>Stateful mode</strong>: The agent retains context and conversation history, allowing for more contextualized interactions. This mode is useful for applications where you need to maintain context across multiple turns. Using stateful mode is recommended for better accuracy and personalized responses.</li></ul><p>Choose a conversation mode based on your application’s requirements for conversation history and context persistence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*P-LksLJ3XIMBMpfb.png" /></figure><blockquote>IAM roles</blockquote><p>IAM roles control who can create, manage, share, and interact with Conversational Analytics API agents. The following table describes the key <a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics-api/access-control">IAM roles for the Conversational Analytics API</a>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/961/1*XOQKK2ZTwm862tixPLIbBA.png" /></figure><blockquote>What’s new with Data Agents:</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*DYedTTJCuKmlQhbTbZjI4g.png" /><figcaption>Image captured from “<a href="https://www.youtube.com/watch?v=Z-AfOcWO_kk">What’s new with data agents</a>”</figcaption></figure><blockquote>Reference articles:</blockquote><p><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics">https://docs.cloud.google.com/gemini/data-agents/conversational-analytics</a></p><p><a href="https://docs.cloud.google.com/gemini/data-agents/conversational-analytics-api/key-concepts">https://docs.cloud.google.com/gemini/data-agents/conversational-analytics-api/key-concepts</a></p><p>​</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=02751397ba97" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Google’s “Trial credit for GenAI App Builder”?]]></title>
            <link>https://medium.com/@skabhianan/what-is-googles-trial-credit-for-genai-app-builder-c94578ad68cd?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c94578ad68cd</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Sun, 31 May 2026 14:49:52 GMT</pubDate>
            <atom:updated>2026-05-31T14:58:11.898Z</atom:updated>
            <content:encoded><![CDATA[<p>There is a $1000 credit 😍​ in my Google paid account named <em>“Trial credit for GenAI App Builder.”</em></p><p>👉​ You can view it in the <em>“Credit”</em> section under the <em>“Billing”</em> option in the GCP Console.</p><p>Of course, these efforts aim to determine how effectively it can be used. 😃​</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UOZr-xOg7F2vq0T60oFNVg.png" /></figure><blockquote>General Google searches for <em>“Google Credits”</em> reveal various credit programs, such as the <em>“Google Cloud Free Trial”</em> with $300 and the <em>“Google for Startups Cloud Program”</em> with $350,000 but we couldn’t find information about this specific credit.</blockquote><p>Then, Gemini App helped me as a handy tool (as usual 🤩​) in getting the relevant information.</p><p>Here are the questions and answers about the <em>“Trial credit for GenAI App Builder”</em>:</p><blockquote>What is the “Trial credit for GenAI App Builder,” and is it offered directly by Google?</blockquote><p>Yes, Google directly offers and issues the “Trial credit for GenAI App Builder,” but it isn’t a standard credit you receive just by signing up for a basic Google Cloud account.</p><p>Instead, it is a highly targeted promotional credit that Google provides through specific developer, startup, and enterprise programs.</p><blockquote>Where does it come from?</blockquote><p>Google distributes this specific credit through a few main official channels:</p><ul><li><strong>The Google Cloud Innovators / Developer Program:</strong> Google regularly runs specialized cohorts, “Fast-Start” initiatives, and multi-day AI hackathons/workshops for developers. When you register for these official events, Google often deposits these targeted credits into your billing account so you can learn their tools without financial risk.</li><li><strong>Google Cloud Startup / Enterprise Programs:</strong> If your company is part of a Google Cloud startup tier or is working with a Google Cloud sales representative to test AI agents, Google will inject this trial credit to fund a Proof of Concept (PoC).</li><li><strong>Vertex AI / GenAI App Builder Launch Promos: </strong>When Google rebranded GenAI App Builder into what is now known as Vertex AI Agent Builder, they distributed specific $1,000 trial packages to cloud administrators to encourage them to try the new “Playbooks” and search grounding features.</li></ul><blockquote>What CAN you use it for?</blockquote><p>It only applies to billing events triggered by <strong>Vertex AI Agent Builder</strong>, which includes:</p><ul><li><strong>Vertex AI Search</strong> (Data store queries, indexing enterprise documents, website search)</li><li><strong>Agent Engine Runtime &amp; Memory</strong> (vCPU-hours and GB-hours used by conversational playbooks)</li><li><strong>Dialogflow CX</strong> (Enterprise conversational layers)</li><li><strong>Grounded Generation API</strong> (RAG-as-a-service)</li></ul><blockquote>How to Use the Credit (Recommended Codelabs)</blockquote><p>To actually consume this credit without accidentally spending out-of-pocket money, you must build applications inside the <strong>Vertex AI Agent Builder</strong> console. The following official Google Codelabs are perfectly tailored for this:</p><blockquote>1. Building AI Agents with Vertex AI Agent Builder</blockquote><p>This is the most direct way to utilize your credits. It guides you through creating a generative travel agent using playbooks, natural language instructions, and grounded data stores.</p><ul><li><strong>What you will do:</strong> Create an Agent, set goals and instructions using Gemini 1.5 Flash, attach a custom data store, and test it in the Agent Studio simulator.</li><li><strong>Link:</strong> <a href="https://codelabs.developers.google.com/devsite/codelabs/building-ai-agents-vertexai">Google Codelabs: Building AI Agents with Vertex AI Agent Builder</a></li></ul><blockquote>2. Building Applications in the AI Era</blockquote><p>This lab teaches you how to orchestrate a data-rich application that leverages Agent Builder’s search capabilities to ground LLM responses with real data.</p><ul><li><strong>What you will do:</strong> Take a massive dataset (like a recipe index), move a subset of it into an environment where Agent Builder can index it, and build a search/chat utility on top of it.</li><li><strong>Link:</strong> <a href="https://codelabs.developers.google.com/codelabs/building-applications-in-the-ai-era">Google Codelabs: Building Applications in the AI Era</a></li></ul><blockquote>Quick Tip to Avoid Surprise Bills:</blockquote><p>When working through these labs, always keep your data stores relatively small (e.g., under 100,000 documents) and <strong>delete your agents, data stores, and conversational search applications as soon as you finish the lab</strong>. While the Agent Builder costs will hit your $1,000 credit, any stray data left sitting in Cloud Storage or BigQuery will count against your standard billing.</p><blockquote>​Why does Google make it separate?</blockquote><p>If you look at your Google Cloud Billing console, you might see this credit listed completely separate from your main “$300 New Customer Free Trial”.</p><p>Google structures it this way on purpose. Because Vertex AI Agent Builder queries (like grounding data with Gemini or indexing enterprise search) can scale up in cost very quickly if you feed them massive corporate databases, Google locks these credits exclusively to SKU Group: GenAI App Builder.</p><p>This allows you to safely experiment with intensive enterprise AI workflows while ensuring that if you accidentally run out of credits, it won’t instantly bleed into and shut down your core infrastructure (like your website’s virtual machines or databases).</p><blockquote>Can you use this credit for general Google Services?</blockquote><p><strong>No.</strong> You cannot use the GenAI App Builder trial credit for core GCP infrastructure services.</p><p>This credit is strictly restricted to <strong>Vertex AI Agent Builder</strong> (formerly GenAI App Builder) ecosystem tools. If you use the services you listed, they will be billed to your standard $300 Google Cloud Free Trial credit or your main payment method instead.</p><p>Here is the breakdown for your list:</p><ul><li><strong>Compute Engine (virtual machines)</strong> — Not covered.</li><li><strong>Cloud Storage (object storage)</strong> — Not covered <em>(Note: While Agent Builder links to Cloud Storage to ingest data, the actual storage cost of hosting files in GCS is not covered by this specific credit).</em></li><li><strong>Cloud SQL (database services)</strong> — Not covered.</li><li><strong>Cloud Functions / Cloud Run (serverless)</strong> — Not covered.</li><li><strong>Kubernetes Engine (GKE)</strong> — Not covered.</li><li><strong>BigQuery (data warehousing)</strong> — Not covered <em>(Note: Like GCS, using BigQuery as a data source for your AI agent is supported, but the BigQuery storage/query costs themselves are not covered).</em></li><li><strong>Firebase</strong> — Not covered.</li><li><strong>Google AI Studio / Partner Models</strong> — Not covered (You cannot use it for third-party models like Claude on Vertex, or standalone Gemini API keys from AI Studio).</li></ul><blockquote>The following Google documentation covers the related billing information:</blockquote><p><a href="https://cloud.google.com/skus/sku-groups/vertex-ai-search-and-conversation">SKU Groups — Vertex AI Search and Conversation</a></p><p><a href="https://cloud.google.com/generative-ai-app-builder/pricing?hl=en">Agent Search pricing</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c94578ad68cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Writing articles publicly — like writing in Medium!!]]></title>
            <link>https://medium.com/@skabhianan/writing-articles-publicly-like-writing-in-medium-83469e04c7d8?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/83469e04c7d8</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Mon, 25 May 2026 08:38:30 GMT</pubDate>
            <atom:updated>2026-05-25T08:39:53.370Z</atom:updated>
            <content:encoded><![CDATA[<h4>The article that I am referring to:<a href="https://medium.com/the-tech-notes/google-cloud-is-doomed-and-aws-knows-it-751d5bb3ce2b">​</a></h4><h3><a href="https://medium.com/the-tech-notes/google-cloud-is-doomed-and-aws-knows-it-751d5bb3ce2b">Google Cloud is Doomed (And AWS Knows It)</a></h3><p>Dear readers,<br>It may sound strange to proclaim the reason for sharing this article publicly, but please excuse me; the truth is:</p><p>I am NOT sharing this article for the actual information the author intended to convey. Interrupt the content of this article on your own along with the 19 comments expressing various views.</p><p>As a beginner at writing articles on Medium, this article helped me understand how important it is to choose the “Title”.</p><p>​And the significance of including a “disclaimer” section in an article.</p><p>For the above referred article, many complained (refer to the comments section of it) that the article is a “clickbait.”</p><p>At the same time, when you need to develop your reader community, it’s important to have a catchy title that encourages readers to go through the article.</p><p>There’s a high chance of counterarguments when you comment publicly, even when supported by statistics.</p><p>Oh, yeah!!, I started learning how to write an article!!</p><p>Last but not least, I personally experienced a steep vocabulary learning curve from this article, thanks to the author for that!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=83469e04c7d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Password Policy for Database user accounts:Part-2]]></title>
            <link>https://medium.com/@skabhianan/password-policy-for-database-user-accounts-part-2-8baa18577aca?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/8baa18577aca</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Fri, 15 May 2026 13:03:12 GMT</pubDate>
            <atom:updated>2026-05-15T13:03:12.082Z</atom:updated>
            <content:encoded><![CDATA[<h4>Recap from Part-1</h4><p>In Part 1 of this article series, we discussed password policy concepts for database user accounts, with a focus on enabling password policies for Cloud SQL instances in the Google Cloud environment.</p><h4>Quick overview on Cloud SQL</h4><p>Cloud SQL is a fully managed, cost-effective relational database service in Google Cloud that automates database setup, maintenance, and administration.</p><p>It is designed to handle structured data and standard SQL operations, so you can focus on application development rather than managing infrastructure.</p><p>Cloud SQL currently supports three major relational database engines: <a href="https://www.mysql.com/">MySQL</a>, <a href="https://www.postgresql.org/">PostgreSQL</a>, and SQL Server</p><p>In this article, we discuss how built-in authentication works on Cloud SQL instances and how database administrators can set password policies for local database users in Cloud SQL for MySQL and PostgreSQL. It is also important to understand where these password policies can be enabled. It makes sense to start with built-in authentication, because that is where these password policies can be applied.</p><h4>Introduction for Built-in Authentication for Cloud SQL instances</h4><p>Authentication is the process of verifying the identity of a user attempting to access an instance. Cloud SQL uses the following types of authentication for database users:</p><p>The database’s built-in authentication uses a username and a password to authenticate local database users.</p><p>IAM database authentication uses IAM to authenticate a user. For more information, see <a href="https://docs.cloud.google.com/sql/docs/mysql/authentication">Overview of Cloud SQL IAM database authentication</a>.</p><p>Although IAM database authentication is more secure and reliable, you might prefer built-in authentication, or a hybrid model that includes both authentication types.</p><p>You can create and manage local database users within a database to allow specific people or applications to access it. These users own the objects they create in the database. Cloud SQL also offers strong built-in password enforcement, which you can define and enable through password policies.</p><h4>Points to consider</h4><ul><li>The password policy for local database users can be enabled when you create the instance, and it can also be enabled later.</li><li>Enabling the password policy for the Cloud SQL instance involves two types of configuration: 1. Instance level 2. Local user account level</li><li>Password policies don’t apply to encrypted passwords.</li><li>If you set user password policies, and you want to deactivate the <a href="https://docs.cloud.google.com/sql/docs/postgres/built-in-authentication#instance_password_policies">password policy for the instance</a>, then you must first remove all user password policies.</li></ul><h3>Cloud SQL for MySQL — Instance password policies:</h3><p><strong>Minimum length:</strong> specify the minimum number of characters that the password must have.</p><p><strong>Password complexity:</strong> check if the password is a combination of lowercase, uppercase, numeric, and non-alphanumeric characters.</p><p><strong>Restrict password reuse:</strong> specify the number of previous passwords that you can’t reuse.</p><p><strong>Disallow username:</strong> prevent the use of the username in the password.</p><h4>Here’s the new console user interface for the Instance Level Policy for MySQL:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1006/0*VFcmBjaFmki11Jle.jpg" /></figure><h3>Cloud SQL for MySQL — User password policies:</h3><p><strong>Set password to expire:</strong> specify the number of days after which the password expires and you need to create a new one.</p><p><strong>Lock after failed attempts:</strong> specify the number of times that you can try the password incorrectly before the account is locked.</p><p><strong>Require current password when password is changed:</strong> require you to enter your existing password when attempting to change it.</p><h4>Here’s the new user interface for the User Level Policy for MySQL:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/779/0*_N2_dRNS6-Fmm0p3.jpg" /></figure><h3>Cloud SQL for PostgreSQL — Instance password policies:</h3><p><strong>Minimum length: </strong>specify the minimum number of characters that the password must have.</p><p><strong>Password complexity:</strong> check if the password is a combination of lowercase, uppercase, numeric, and non-alphanumeric characters.</p><p><strong>Restrict password reuse:</strong> specify the number of previous passwords that you can’t reuse.</p><p><strong>Disallow username:</strong> prevent the use of the username in the password.</p><p><strong>Set password change interval:</strong> specify the minimum duration after which you can change the password.</p><h4>Here’s the new console user interface for the Instance Level Policy for PostgreSQL:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/764/0*v0AWXc2Clrm0JbbV.jpg" /></figure><h3>Cloud SQL for PostgreSQL — User password policies:</h3><p><strong>Set password to expire:</strong> specify the number of days after which the password expires and you need to create a new one.</p><p><strong>Lock after failed attempts:</strong> specify the number of times that you can try the password incorrectly before the account is locked.</p><p><strong>Note:</strong></p><p>Unlike Cloud SQL for MySQL, there is no console interface available for applying the User password policies for Cloud SQL for PostgreSQL instance.</p><p>For applying the “Set password to expire” and “Lock after failed attempts” policies, we need to use the following gcloud command:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/792/1*tprCIHaGLIAJsXoZTtsrzA.png" /></figure><p>To set the user password policy, use the <a href="https://docs.cloud.google.com/sdk/gcloud/reference/sql/users/set-password-policy">gcloud sql users set-password-policy</a> command.</p><p>Replace the following:</p><p>USER_NAME: The user name.</p><p>INSTANCE_NAME: The name of the instance.</p><p>HOST: The user’s host name as a specific IP address, address range, or any host (%).</p><p>PASSWORD_POLICY_ALLOWED_FAILED_ATTEMPTS: Optional: The number of times that a user can try the password incorrectly before the account is locked. Use -password-policy-enable-failed-attempts-check to enable and -no-password-policy-enable-failed-attempts-check to disable the check.</p><p>PASSWORD_POLICY_PASSWORD_EXPIRATION_DURATION: Optional: Specifies the number of days after which the password expires and the user needs to create a new one.</p><p>In the next part of this article series, we will discuss configuring alert for password expiry and alert for user account lock upon the configured unsuccessful attempts.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8baa18577aca" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Password Policy for Database user accounts:Part-1]]></title>
            <link>https://medium.com/@skabhianan/password-policy-for-database-user-accounts-part-1-964b0526313d?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/964b0526313d</guid>
            <category><![CDATA[policy]]></category>
            <category><![CDATA[compliance]]></category>
            <category><![CDATA[strong-password-policy]]></category>
            <category><![CDATA[database]]></category>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Sat, 25 Apr 2026 19:07:21 GMT</pubDate>
            <atom:updated>2026-04-25T19:07:21.064Z</atom:updated>
            <content:encoded><![CDATA[<p>This article series discusses password policies for database user accounts, with a focus on managed databases in the GCP environment, especially CloudSQL for MySQL and CloudSQL for Postgres. Before diving in, it is helpful to cover the following password policy topics to understand their significance.</p><h3>What is password policy?</h3><p>A password policy is a clear set of rules that explains how to create, use, and protect passwords at work. It helps reduce mistakes that often lead to unauthorized access or security incidents.</p><h3>The significance of password policy:</h3><p>A strong password policy can help you prevent security threats and reduce the risk of cyberattacks. Implementing an effective password policy and MFA can also help you comply with regulations like HIPAA, the NIST’s Cybersecurity Framework (CSF), the GDPR, SOC2, the PCI DSS, and NIS2.</p><p>Hackers increasingly use weak and stolen credentials to compromise business-critical environments. Gaining access with a known password for a user account is often much easier than exploiting other vulnerabilities. Therefore, strong password security and robust password policies are an excellent way for organizations to bolster their cybersecurity posture.</p><h3>Why we needed password policy &amp; compliance?</h3><p>Password policies and compliance are rules and methods that enforce the user for using a secure and robust password. A billion credentials were stolen last year from multiple data breaches. According to <a href="https://enterprise.verizon.com/resources/reports/2017_dbir.pdf">Verizon’s Data Breach Report</a>, 81% of data breaches are caused by compromised, weak, and reused passwords.</p><h3>What characteristics make up an effective password policy?</h3><p>Here are some best practice guidelines for effective password policies:</p><p>Encourage the use of passphrases</p><p>Don’t throw away password expiry</p><p>Implement breached password protection</p><p>Use password dictionary checks</p><p>Use account lockout policies</p><h3>Legally required password constraints:</h3><p>Many organizations are legally required to enforce strict password constraints. These requirements are based on extensive study of the mathematical principles of password entropy and are proven to create stronger passwords. They combine a variety of strategies, including:</p><p>Minimum and maximum character counts</p><p>Uppercase and lowercase letter requirements</p><p>Numeric and special character requirements</p><p>Time-based password resets</p><p>Password re-use policies</p><h3>Password policies to improve password strength:</h3><h4>1. Minimum Password Age policy</h4><p>The Minimum Password Age policy defines the minimum number of days users must keep a password before they can change it.</p><h4>2. Enforce Password History policy</h4><p>The “Enforce password history” policy ensures users must create a set number of unique passwords before they can reuse an old one. This is important because password reuse is common, since users often feel more comfortable with familiar passwords. Reusing the same password for a long time increases the risk of compromise, such as through a brute force attack. A password age policy is not effective without a password history policy. If users can change their password but reuse an old one, the effectiveness of a password age policy is greatly reduced.</p><h4>3. Minimum Password Length policy</h4><p>The Minimum Password Length policy defines the minimum number of characters required to create a password. Minimum Password Length should be at least eight characters. Longer passwords are generally more secure and harder to crack than shorter ones. For even greater security, you could set the minimum password length to 14 characters.</p><h4>4. Passwords Must Meet Complexity Requirements policy</h4><p>The Password Complexity Requirements policy ensures users do not use basic passwords. Passwords should be a combination of uppercase and lowercase letters, numbers, and some special characters. We can set the following policies in the Password Complexity Requirements:</p><p>Passwords shouldn’t contain the username or basic profile fields, such as a first name.</p><p>The password must use the following character combinations:</p><p>Uppercase letters</p><p>Lowercase letters</p><p>Non-alphanumeric characters</p><p>(special characters): (~!@#$%^&amp;*_-+=`|(){}[]:;”’&lt;&gt;,.?/)</p><p>Numeric characters</p><h4>5. Common Password Protection</h4><p>Users should not use common passwords, so restrict the use of common passwords.</p><h4>6. Dictionary Password Prevention</h4><p>A password dictionary is a file that contains a list of potential passwords. This feature prevents users from setting a password found in the dynamic password dictionary. We use this dynamic password dictionary to prevent the use of dictionary passwords.</p><h4>7. Password Audit policy</h4><p>Enabling the Password Audit policy lets you track all password changes. Monitoring these changes makes it easier to identify potential security issues. This helps ensure user accountability and provides evidence in the event of a security breach.</p><p>In the next part of this article series, we will continue discussing password policy configuration for CloudSQL for MySQL and CloudSQL for Postgres instances.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=964b0526313d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Claude Helped Me Create an Automated CloudSQL PostgreSQL Backup Script]]></title>
            <link>https://medium.com/@skabhianan/how-claude-helped-me-create-an-automated-cloudsql-postgresql-backup-script-2ce70f70fd4a?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/2ce70f70fd4a</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Wed, 02 Jul 2025 18:32:19 GMT</pubDate>
            <atom:updated>2025-07-02T18:32:19.651Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>As a Database Administrator (DBA) working with Google Cloud Platform, I frequently needed to backup specific databases from CloudSQL PostgreSQL instances and restore them to different instances. This process, when done manually, was time-consuming and error-prone. I decided to leverage Claude’s capabilities to help me create a comprehensive automation script that would streamline this workflow.</p><h3>The Problem Statement</h3><p>I needed a solution that could:</p><ul><li>Backup a specific database (not the entire instance) from a CloudSQL PostgreSQL instance</li><li>Restore that backup to a completely different CloudSQL instance</li><li>Handle cross-project scenarios</li><li>Provide safety checks and error handling</li><li>Be reusable and maintainable</li></ul><h3>My Interaction with Claude</h3><h3>The Question I Asked</h3><p>I approached Claude with a specific request:</p><blockquote><em>“In the CloudSQL Postgres Automation script, please modify it to take the backup of a specific database hosted in a cloudsql instance and the backup has to be restored on another instance.”</em></blockquote><h3>Claude’s Response</h3><p>Claude immediately understood the requirements and provided a comprehensive solution. Instead of just giving me basic commands, Claude created a full-featured bash script with extensive functionality.</p><h3>Real-World Challenge: Permission Issues</h3><p>During the initial implementation, I encountered a common CloudSQL permission error:</p><pre>ERROR: (gcloud.sql.import.sql) HTTPError 412: The service account does not have the required permissions for the bucket.</pre><p>When I brought this issue back to Claude, the AI immediately understood the problem and provided an enhanced solution.</p><h3>The Follow-up Question</h3><blockquote><em>“while running this cloudsql postgresql backup script getting the following error: ERROR: (gcloud.sql.import.sql) HTTPError 412: The service account does not have the required permissions for the bucket.”</em></blockquote><h3>Claude’s Enhanced Solution</h3><p>Claude didn’t just provide a quick fix — it delivered a comprehensive enhancement that:</p><ol><li><strong>Automatically detected CloudSQL service accounts</strong> for both source and target instances</li><li><strong>Granted appropriate permissions</strong> automatically during script execution</li><li><strong>Added validation checks</strong> to prevent the error from occurring</li><li><strong>Included troubleshooting documentation</strong> for manual resolution if needed</li></ol><h3>Enhanced Script Features (Post-Issue Resolution)</h3><h3>Complete Script Features</h3><p>Claude provided a robust automation script with the following capabilities:</p><h4>1. Automated Permission Management</h4><ul><li>Automatic detection of CloudSQL service accounts</li><li>Dynamic permission granting to GCS buckets</li><li>Support for cross-project service account permissions</li><li>Built-in validation to prevent permission errors</li></ul><h4>2. Configuration Management</h4><ul><li>Easy-to-configure variables at the top of the script</li><li>Support for different source and target projects</li><li>Flexible naming conventions for backups</li><li>Configurable regions and storage locations</li></ul><h4>3. Safety and Validation</h4><ul><li>Prerequisites validation (gcloud CLI, gsutil, authentication)</li><li>GCS bucket existence verification</li><li>Configuration completeness checks</li><li>User confirmation before destructive operations</li><li>Comprehensive error handling with exit codes</li></ul><h4>4. User Experience</h4><ul><li>Colored output for different message types (info, success, warning, error)</li><li>Clear progress indicators</li><li>Interactive prompts for critical decisions</li><li>Detailed help documentation</li><li>Built-in troubleshooting guidance</li></ul><h4>5. Operational Features</h4><ul><li>Database-specific backup (not full instance)</li><li>Cross-instance restoration capability</li><li>Cross-project support</li><li>Automatic cleanup options</li><li>Verification steps</li></ul><h4>6. Advanced Functionality</h4><ul><li>Dry-run mode for testing</li><li>Timestamped backup files</li><li>Operation status monitoring</li><li>Connection information for verification</li><li>Automatic service account permission management</li></ul><h3>Key Features Breakdown</h3><h3>1. Automated Service Account Permission Management</h3><pre># Get CloudSQL service accounts<br>SOURCE_SA=$(get_cloudsql_service_account $SOURCE_PROJECT_ID $SOURCE_INSTANCE_ID)<br>TARGET_SA=$(get_cloudsql_service_account $TARGET_PROJECT_ID $TARGET_INSTANCE_ID)</pre><pre># Grant permissions automatically<br>gsutil iam ch serviceAccount:$SOURCE_SA:objectAdmin gs://$BACKUP_BUCKET<br>gsutil iam ch serviceAccount:$TARGET_SA:objectAdmin gs://$BACKUP_BUCKET</pre><p>This was the game-changer — automatically handling the most common failure point.</p><h3>2. Database-Specific Backup</h3><pre>gcloud sql export sql $SOURCE_INSTANCE_ID $BACKUP_URI \<br>    --database=$SOURCE_DATABASE_NAME \<br>    --project=$SOURCE_PROJECT_ID \<br>    --quiet</pre><p>Unlike full instance backups, this targets only the specific database you need.</p><h3>3. Cross-Instance Restoration</h3><pre>gcloud sql import sql $TARGET_INSTANCE_ID $BACKUP_URI \<br>    --database=$TARGET_DATABASE_NAME \<br>    --project=$TARGET_PROJECT_ID \<br>    --quiet</pre><p>Seamlessly restores to a different CloudSQL instance, even in different projects.</p><h3>4. Intelligent Database Management</h3><p>The script handles target database preparation:</p><ul><li>Checks if target database exists</li><li>Offers to drop and recreate if needed</li><li>Creates new database if it doesn’t exist</li></ul><h3>5. Enhanced Error Handling and Validation</h3><pre>set -e  # Exit on any error</pre><pre># Validate bucket existence<br>if ! gsutil ls gs://$BACKUP_BUCKET &amp;&gt; /dev/null; then<br>    print_error &quot;GCS bucket &#39;gs://$BACKUP_BUCKET&#39; does not exist or is not accessible&quot;<br>    exit 1<br>fi</pre><pre>if [ $? -eq 0 ]; then<br>    print_success &quot;Operation completed successfully&quot;<br>else<br>    print_error &quot;Operation failed&quot;<br>    exit 1<br>fi</pre><h3>6. User-Friendly Interface</h3><ul><li>Color-coded output messages</li><li>Interactive confirmations</li><li>Clear progress indicators</li><li>Detailed help system</li><li>Built-in troubleshooting guidance</li></ul><h3>How the Script Works</h3><h3>Step-by-Step Process</h3><p><strong>Validation Phase</strong></p><ul><li>Checks for gcloud CLI and gsutil installation</li><li>Validates authentication status</li><li>Verifies GCS bucket existence and accessibility</li><li>Ensures all configuration variables are set</li><li>Confirms required permissions</li></ul><p><strong>Permission Setup</strong></p><ul><li>Automatically detects CloudSQL service accounts for both instances</li><li>Grants Storage Object Admin role to source instance service account</li><li>Grants Storage Object Admin role to target instance service account</li><li>Validates permission grants</li></ul><p><strong>Backup Creation</strong> 3. <strong>Backup Creation</strong></p><ul><li>Sets the source project context</li><li>Exports the specific database to Google Cloud Storage</li><li>Generates timestamped backup files</li><li>Monitors operation completion</li></ul><p><strong>Target Preparation</strong></p><ul><li>Switches to target project context</li><li>Handles existing database scenarios</li><li>Creates or recreates target database</li><li>Ensures clean restoration environment</li></ul><p><strong>Restoration Process</strong></p><ul><li>Imports backup from GCS to target instance</li><li>Monitors import operation status</li><li>Provides success confirmation</li><li>Offers verification guidance</li></ul><p><strong>Cleanup and Verification</strong></p><ul><li>Optional cleanup of temporary backup files</li><li>Provides connection information</li><li>Suggests verification steps</li></ul><h3>The Permission Challenge Solution</h3><p>The most significant enhancement came from addressing a real-world issue. The original script worked in theory but failed in practice due to CloudSQL service account permissions. Claude’s solution included:</p><h3>Automatic Service Account Detection</h3><pre>get_cloudsql_service_account() {<br>    local project_id=$1<br>    local instance_id=$2<br>    <br>    gcloud sql instances describe $instance_id --project=$project_id --format=&quot;value(serviceAccountEmailAddress)&quot;<br>}</pre><h3>Dynamic Permission Granting</h3><pre># Grant permissions to both source and target service accounts<br>gsutil iam ch serviceAccount:$SOURCE_SA:objectAdmin gs://$BACKUP_BUCKET<br>gsutil iam ch serviceAccount:$TARGET_SA:objectAdmin gs://$BACKUP_BUCKET</pre><h3>Built-in Troubleshooting</h3><p>The script now includes comprehensive troubleshooting documentation and manual fix commands for edge cases.</p><h3>Configuration Example</h3><pre># Source Configuration<br>SOURCE_PROJECT_ID=&quot;my-source-project&quot;<br>SOURCE_INSTANCE_ID=&quot;source-postgres-instance&quot;<br>SOURCE_DATABASE_NAME=&quot;production_db&quot;<br>SOURCE_USER=&quot;postgres&quot;</pre><pre># Target Configuration<br>TARGET_PROJECT_ID=&quot;my-target-project&quot;<br>TARGET_INSTANCE_ID=&quot;target-postgres-instance&quot;<br>TARGET_DATABASE_NAME=&quot;restored_production_db&quot;<br>TARGET_USER=&quot;postgres&quot;</pre><pre># Backup Configuration<br>BACKUP_BUCKET=&quot;my-backup-bucket&quot;<br>BACKUP_FILE_PREFIX=&quot;db-backup&quot;<br>REGION=&quot;us-central1&quot;</pre><h3>Usage Scenarios</h3><h3>Basic Usage</h3><pre>./backup_restore.sh</pre><h3>Dry Run (Testing)</h3><pre>./backup_restore.sh --dry-run</pre><h3>Getting Help</h3><pre>./backup_restore.sh --help</pre><h3>What Made Claude’s Solution Exceptional</h3><h3>1. Comprehensive Approach</h3><p>Claude didn’t just provide the basic gcloud commands. Instead, it delivered a production-ready script with enterprise-level features.</p><h3>2. Best Practices Implementation</h3><ul><li>Error handling with set -e</li><li>Colored output for better UX</li><li>Interactive confirmations for safety</li><li>Comprehensive logging</li></ul><h3>3. Real-World Considerations</h3><ul><li>Cross-project scenarios</li><li>Existing database handling</li><li>Cleanup procedures</li><li>Verification steps</li></ul><h3>4. Documentation and Usability</h3><ul><li>Built-in help system</li><li>Clear configuration instructions</li><li>Usage examples</li><li>Prerequisites checklist</li></ul><h3>5. Flexibility and Extensibility</h3><h3>6. Documentation and Usability</h3><ul><li>Built-in help system</li><li>Clear configuration instructions</li><li>Usage examples</li><li>Prerequisites checklist</li><li>Troubleshooting guide with manual fixes</li></ul><h3>7. Flexibility and Extensibility</h3><ul><li>Configurable variables</li><li>Modular functions</li><li>Easy to customize</li><li>Dry-run capability</li><li>Automatic permission management</li></ul><h3>Prerequisites for Using the Script</h3><h3>Technical Requirements</h3><ul><li><strong>gcloud CLI</strong>: Installed and authenticated</li><li><strong>gsutil</strong>: Included with Google Cloud SDK</li><li><strong>IAM Permissions</strong>:</li><li>Cloud SQL Admin role</li><li>Storage Admin role for GCS bucket</li><li>IAM Admin role (for granting service account permissions)</li><li><strong>GCS Bucket</strong>: For temporary backup storage</li><li><strong>Network Access</strong>: Between instances if needed</li></ul><h3>Access Requirements</h3><ul><li>Access to both source and target GCP projects</li><li>Appropriate database permissions</li><li>GCS bucket read/write permissions</li></ul><h3>Benefits Achieved</h3><ol><li><strong>Time Savings</strong>: Automated a previously manual process</li><li><strong>Error Reduction</strong>: Built-in validation and error handling</li><li><strong>Consistency</strong>: Standardized backup and restore procedures</li><li><strong>Scalability</strong>: Easy to use across multiple projects and instances</li><li><strong>Safety</strong>: Multiple confirmation steps and dry-run capability</li><li><strong>Maintainability</strong>: Well-documented and modular code</li></ol><h3>Real-World Impact</h3><p>This script has transformed our database management workflow:</p><ul><li><strong>Reduced backup/restore time</strong> from 30+ minutes to 5 minutes</li><li><strong>Eliminated human errors</strong> in manual database operations</li><li><strong>Solved the most common CloudSQL backup failure</strong> (service account permissions)</li><li><strong>Enabled consistent database procedures</strong> across different environments</li><li><strong>Improved disaster recovery capabilities</strong></li><li><strong>Facilitated database environment provisioning for development teams</strong></li><li><strong>Removed the need for manual permission management</strong></li></ul><h3>Lessons Learned</h3><h3>Working with Claude</h3><ol><li><strong>Be Specific</strong>: Clear requirements lead to better solutions</li><li><strong>Report Real Issues</strong>: Claude excels at solving actual problems encountered in production</li><li><strong>Ask for Production-Ready Code</strong>: Claude can deliver enterprise-level scripts</li><li><strong>Expect Iterative Improvement</strong>: Claude enhances solutions when presented with real-world challenges</li><li><strong>Request Documentation</strong>: Claude provides excellent inline documentation and troubleshooting guides</li><li><strong>Leverage Best Practices</strong>: Claude incorporates industry standards automatically</li></ol><h3>Script Development</h3><ol><li><strong>Error Handling is Crucial</strong>: The script’s robustness comes from comprehensive error handling</li><li><strong>Permission Management is Key</strong>: Automatic service account permission handling prevents the most common failures</li><li><strong>User Experience Matters</strong>: Colored output and clear messages make a huge difference</li><li><strong>Safety First</strong>: Interactive confirmations prevent costly mistakes</li><li><strong>Validation Prevents Issues</strong>: Comprehensive upfront validation saves time and prevents failures</li><li><strong>Documentation Saves Time</strong>: Built-in help and troubleshooting reduce support overhead</li></ol><h3>Future Enhancements</h3><p>Based on the solid foundation Claude provided, potential enhancements include:</p><ul><li><strong>Scheduling capabilities</strong> with cron integration</li><li><strong>Slack/email notifications</strong> for operation status</li><li><strong>Multiple database backup</strong> in single operation</li><li><strong>Backup retention policies</strong></li><li><strong>Integration with monitoring systems</strong></li><li><strong>Configuration file support</strong> instead of script variables</li></ul><h3>Conclusion</h3><p>Claude’s assistance in creating this CloudSQL backup automation script demonstrates the power of AI-assisted database administration. What could have taken days of research, testing, and refinement was accomplished in a single interaction. The resulting script is not just functional but production-ready, incorporating best practices and real-world considerations that might have been overlooked in manual development.</p><p>The script has become an essential tool in our database administration toolkit, saving time, reducing errors, and improving our overall database management capabilities. This experience reinforces the value of leveraging AI tools like Claude for database automation and administration challenges.</p><p>Here we go for the complete script published in Claude itself:</p><p><a href="https://claude.ai/public/artifacts/a090cf99-3364-470b-9d0b-464b4492e85d">https://claude.ai/public/artifacts/a090cf99-3364-470b-9d0b-464b4492e85d</a></p><h3>About the Author</h3><p>This article documents a real-world experience of using Claude to solve a specific database administration challenge. The resulting automation script is now used in production environments and has been shared with the broader DBA community.</p><p><em>If you found this article helpful, consider sharing your own experiences with AI-assisted database administration. The combination of human database expertise and AI capabilities can produce remarkable results in database automation and management.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2ce70f70fd4a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Invoking Visual Studio Code in Google Cloud Shell and running Terraform script locally in a GCP…]]></title>
            <link>https://medium.com/@skabhianan/invoking-visual-studio-code-in-google-cloud-shell-and-running-terraform-script-locally-in-a-gcp-8ceda029f7b1?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/8ceda029f7b1</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Sun, 10 Mar 2024 18:16:07 GMT</pubDate>
            <atom:updated>2024-03-10T18:16:07.650Z</atom:updated>
            <content:encoded><![CDATA[<h3>Invoking Visual Studio Code in Google Cloud Shell and running Terraform script locally in a GCP Project</h3><blockquote>With respect to Security perspective, there is no question that we have to follow the standard best practices like Workload Identity Federation. This requires a few efforts and understanding and often times involves multiple teams involvement in a typical working environment.</blockquote><p>However, in a scenario in which you have limited/restricted IAM permissions on a GCP Project and required to implement/install something quickly in a faster way, for example a Data Engineer who has only Cloud Admin permisisons wanted to execute a Terrafrom script to create a CloudSQL Intance. In this scenerio, as far as I know this would be the ideal solution.</p><p>Reference article:</p><p><a href="https://medium.com/google-cloud/how-to-run-visual-studio-code-in-google-cloud-shell-354d125d5748">How to run Visual Studio Code in Google Cloud Shell</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L3h5CKzb1sdPapSUBEw3Aw.png" /><figcaption>Install code-server: curl -fsSL <a href="https://code-server.dev/install.sh">https://code-server.dev/install.sh</a> | sh</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/979/1*kGS7jodOKU39_-mbhUEisA.png" /><figcaption>code-server --port 8080</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Cll_JhVOtmWCQns5te8WIA.png" /><figcaption>Click on Web Preview -&gt; Preview on port 8080</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/661/1*oehat-GuHxBM8cJoN8g9HA.png" /><figcaption>Copy the password from config.yaml from the mentioned path</figcaption></figure><h4><strong>That’s it, now you have an IDE running in your browser!</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*8asb1Ht-yVoR1EMRhz_OlA.png" /><figcaption>Terraform installed and used from CloudShell</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZaKaVuzydUfc5t_Gm48EHw.png" /></figure><h4>Not only running the Terrafrom script, all the tasks that can be done from the Visual Studio Code is possible!!! locally to the GCP Project accordingly to your limited/existing Permissions.</h4><p>Any comments, suggestions are most welcome…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8ceda029f7b1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Visual Inspection AI/ML — GCP Vertex AI AutoML]]></title>
            <link>https://medium.com/@skabhianan/visual-inspection-ai-ml-gcp-vertex-ai-automl-e7b89e227220?source=rss-2b89a0dd3bf9------2</link>
            <guid isPermaLink="false">https://medium.com/p/e7b89e227220</guid>
            <dc:creator><![CDATA[Anandan]]></dc:creator>
            <pubDate>Sat, 09 Mar 2024 09:17:20 GMT</pubDate>
            <atom:updated>2024-03-09T09:17:20.510Z</atom:updated>
            <content:encoded><![CDATA[<h3>Visual Inspection AI/ML — GCP Vertex AI AutoML</h3><h4>Inspired by the video tutorial:</h4><p><a href="https://www.youtube.com/watch?v=TFaR4x8BZG8">https://www.youtube.com/watch?v=TFaR4x8BZG8</a></p><p>It’s an excellent, step-by-step video tutorial walk-through for the deployment with neat explanations.</p><p>This article is to shed some light on the Involved Cost, so that if you try to explore on your personal GCP account, you have a fair idea on the same.</p><p>Here we go for the screen-shots of the exploration on my personal GCP account. The detailed steps are available with the referred video.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/776/1*lv_sJQAduxjmPrjwIrbb1Q.png" /><figcaption>Step-1: Training Method</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*IlvB7b5Htf-KOMQ1MI0ImA.png" /><figcaption>Step-2: Model details</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/783/1*4CSk34OGvCOdnlRuJqd_RQ.png" /><figcaption>Step-2 b. Model details continued</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*knJUwI0sSo0nhnyfHv0SWA.png" /><figcaption>Step 3: Training options</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*H3vt8ouO91Kva2ADtpGDEg.png" /><figcaption>Step 4: Compute and pricing</figcaption></figure><p>Be noted for the invovled cost. 1 node hour = 92 n1-standard4 VM instances.</p><p>The following are the excerpts from the Google Vertex AI pricing documentation:</p><p><a href="https://cloud.google.com/vertex-ai/pricing">https://cloud.google.com/vertex-ai/pricing</a></p><h4>Pricing for AutoML models</h4><p>For Vertex AI AutoML models, you pay for three main activities:</p><ul><li>Training the model</li><li>Deploying the model to an endpoint</li><li>Using the model to make predictions</li></ul><p>Vertex AI uses predefined machine configurations for Vertex AutoML models, and the hourly rate for these activities reflects the resource usage.</p><p>The time required to train your model depends on the size and complexity of your training data. Models must be deployed before they can provide online predictions or online explanations.</p><p>You pay for each model deployed to an endpoint, even if no prediction is made. You must undeploy your model to stop incurring further charges. Models that are not deployed or have failed to deploy are not charged.</p><p>You pay only for compute hours used; if training fails for any reason other than a user-initiated cancellation, you are not billed for the time. You are charged for training time if you cancel the operation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/779/1*4O41eAPkBPp56zfHjo6AyA.png" /><figcaption>The images are uploaded to the Google storage bucket (These images are cloned from the shared repository from the referred video tutorial. )</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/783/1*uRTS7nEnuJrJuGd5qhPX-A.png" /><figcaption>The model is being trained</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/783/1*2uPlEXEfYF7mxSr-Y01CBQ.png" /><figcaption>It took 1 hr 43 mins to train the model with 30 images. We used the n1-standard-4 VM type.</figcaption></figure><p>Of course, the training speed can be optimized with powerful instances. The more powerful instance that is being used the more cost involved. (Refer to the pricing page for Google Vertex AI)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/777/1*Ni67K4e-hPVkrQlHeCQAkQ.png" /><figcaption>Step to Deploy the model to an endpoint</figcaption></figure><p>Endpoints are machine learning models made available for online prediction requests. Endpoints are useful for timely predictions from many users (for example, in response to an application request). You can also request batch predictions if you don’t need immediate results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*iuSyMvDjOjHrJZa8txLb2Q.png" /><figcaption>Depoly to endpoint cont’d.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/782/1*NPzAniUH6TNHGLCzhqTfDw.png" /><figcaption>Deploy to endpoint cont’d.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*dg1KKsIhmRvRrqmArN3MSQ.png" /><figcaption>The model is being deployed to an endpoint</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/784/1*ltKD7wWeajvloiAquZCSMQ.png" /><figcaption>And the final step: Uploading an image to the model for the prediction. (The sample images are also available in the repository)</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e7b89e227220" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>