<?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 Akankshabhagwanani on Medium]]></title>
        <description><![CDATA[Stories by Akankshabhagwanani on Medium]]></description>
        <link>https://medium.com/@akankshabhagwanani?source=rss-a93648d4a988------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*azpsenoaNJC4wrNC21LT8A@2x.jpeg</url>
            <title>Stories by Akankshabhagwanani on Medium</title>
            <link>https://medium.com/@akankshabhagwanani?source=rss-a93648d4a988------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 20:50:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@akankshabhagwanani/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[The Agentic Data Cloud Part 2: Activating Data with the Data Agent Kit & Proactive Workflows]]></title>
            <link>https://medium.com/@akankshabhagwanani/the-agentic-data-cloud-part-2-activating-data-with-the-data-agent-kit-proactive-workflows-cd4885a9914d?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd4885a9914d</guid>
            <category><![CDATA[agentic-data-cloud]]></category>
            <category><![CDATA[bigquery]]></category>
            <category><![CDATA[google-cloud-next]]></category>
            <category><![CDATA[data-agents]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Fri, 01 May 2026 15:22:39 GMT</pubDate>
            <atom:updated>2026-05-01T15:22:39.778Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fJTi71u03W9Vn3GeGd0U3A.png" /></figure><p>In <a href="https://medium.com/@akankshabhagwanani/the-agentic-data-cloud-part-1-universal-context-via-knowledge-catalog-ca4b57dbab77">Part 1</a>, we explored how the <strong>Knowledge Catalog</strong> provides the “Universal Context” (the brain) for the Agentic Data Cloud. But context without execution is just information. To move from an agent that simply <em>knows</em> to an agent that <em>does</em>, we need a <strong>System of Action.</strong></p><p>At <strong>Google Cloud Next </strong>’26, we unveiled the framework and the specialized intelligence that turn BigQuery from a data warehouse into an active collaborator: the <strong>Data Agent Kit</strong> and <strong>Proactive Agentic Workflows.</strong></p><h3>The Data Agent Kit: The Engine of Autonomy</h3><p>The <strong>Data Agent Kit</strong> is the connective tissue. It provides a standardized set of “skills” and plugins that allow Gemini to interact directly with your data stack. Instead of building agents from scratch, the Kit allows you to:</p><p><strong>Orchestrate Multi-Step Workflows:</strong> Agents can now move beyond simple Q&amp;A. They can reason across multiple datasets, trigger Vertex AI pipelines, and call external APIs via zero-copy federation.</p><p><strong>Leverage High-Performance Skills:</strong> The kit includes pre-built “agentic skills” — like advanced data profiling and semantic reasoning — that scale to petabytes without moving your data.</p><p><strong>BigQuery Agent Analytics:</strong> For the first time, every action an agent takes is logged back into BigQuery. This creates a transparent audit trail for troubleshooting and optimizing agent “logic” in production.</p><h3>Moving from Reactive to Proactive</h3><p>While we often think of agents as things we “talk to,” Next ’26 introduced a shift toward agents that work <em>for</em> you in the background:</p><h4>1. Proactive Agentic Workflows</h4><p>Leveraging the new <strong>long-running agents</strong> capability, you can now deploy agents that don’t wait for a prompt. These agents monitor your <strong>BigQuery Measures</strong> for anomalies or shifts in KPIs. When they detect a change — like a sudden dip in regional supply chain efficiency — they autonomously perform root-cause analysis and deliver a summarized briefing to your <strong>Agent Inbox</strong> before you’ve even started your day.</p><h4>2. The Deep Research Agent</h4><p>Powered by the <strong>Knowledge Catalog</strong>, this agent represents the evolution of the data workspace. Within Gemini Enterprise and BigQuery, it performs autonomous “Deep Research” by synthesizing data across the entire Agentic Data Cloud. It connects structured BigQuery tables with unstructured documents (like PDF contracts or research papers) in Cloud Storage, providing a holistic answer to complex strategic questions that previously required weeks of manual cross-referencing.</p><h4>3. Conversational Analytics in BigQuery Studio</h4><p>For the hands-on practitioner, BigQuery Studio now features <strong>Conversational Analytics</strong>. This allows you to explore data using natural language while staying within the “semantic guardrails” defined in the Knowledge Catalog. It ensures that when you ask for “Gross Margin,” the agent uses the exact enterprise-verified calculation every time.</p><h3>Moving at the Speed of Action</h3><p>The goal of the Agentic Data Cloud isn’t just to talk about data; it’s to act on it. By combining the <strong>Data Agent Kit</strong> with these proactive capabilities, we are removing the friction between insight and execution.</p><blockquote><strong><em>We’ve moved from a world where you go to the data, to a world where the data — through agents — comes to you with the answers.</em></strong></blockquote><p><strong>Next up in this series:</strong> We’ll dive into the <strong>Open Lakehouse</strong> — the borderless foundation built on Apache Iceberg that allows these agents to work across clouds without the “walled gardens.”</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd4885a9914d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Agentic Data Cloud Part 1: Universal Context via Knowledge Catalog]]></title>
            <link>https://medium.com/@akankshabhagwanani/the-agentic-data-cloud-part-1-universal-context-via-knowledge-catalog-ca4b57dbab77?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/ca4b57dbab77</guid>
            <category><![CDATA[google-cloud-next26]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[knowledge-catalog]]></category>
            <category><![CDATA[agentic-data-cloud]]></category>
            <category><![CDATA[bigquery]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Tue, 28 Apr 2026 13:38:28 GMT</pubDate>
            <atom:updated>2026-04-28T13:38:28.984Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CBmbOnoe1VrFhNopvweWzw.png" /></figure><p>We’ve all heard the pitch: “AI will transform your business.” But here’s the reality — an agent is only as good as the context it’s given. If your data is a messy basement of siloed tables and undocumented PDFs, your “intelligent” agent is basically just guessing.</p><p>At <strong>Google Cloud Next ’26</strong>, the shift from <strong>Generative AI</strong> to the <strong>Agentic Data Cloud</strong> became official. And it all starts with one foundational pillar: <strong>Universal Context.</strong></p><p>To build this, we’re moving past static metadata. We’re moving into the <strong>Knowledge Catalog</strong>.</p><h3>The Architecture of Truth: Aggregation, Enrichment, Search</h3><p>The Knowledge Catalog (the evolution of Dataplex) isn’t just a library; it’s a reasoning engine. It operates on three distinct levels to ensure your agents don’t just “see” data, but actually <em>understand</em> it.</p><h4>1. Aggregation: Unifying the “Where”</h4><p>Context shouldn’t have borders. Knowledge Catalog now aggregates native context across:</p><ul><li><strong>Google Native:</strong> BigQuery, Spanner, and Looker.</li><li><strong>The Ecosystem:</strong> Direct federation with Salesforce Data360, SAP, and third-party metadata connectors.</li><li><strong>The Logic:</strong> With the new <strong>BigQuery Measures</strong>, the catalog understands your KPIs and business semantics. It ensures your agent’s definition of “Gross Margin” matches your Business’s definition.</li></ul><h4>2. Enrichment: Defining the “What”</h4><p>Data without meaning is just noise. The catalog now uses Gemini to perform <strong>Continuous Enrichment</strong>:</p><ul><li><strong>Unstructured Mastery:</strong> The moment a file hits Cloud Storage, <strong>Smart Storage</strong> APIs auto-tag and embed it, tapping into context from unstructured data.</li><li><strong>Entity Mapping:</strong> It doesn’t just index a PDF; it extracts the relationships between entities inside it and maps them back to your structured SQL tables.</li><li><strong>Golden Queries:</strong> It identifies “verified SQL patterns,” providing semantic guardrails so agents don’t hallucinate a join that doesn’t exist.</li></ul><h4>3. Search: Enabling the “How”</h4><p>For an agent, search is the new query path. The Knowledge Catalog uses a hybrid search stack — leveraging decades of Google Search innovation — to deliver <strong>sub-second latency</strong>.</p><ul><li><strong>Precision:</strong> Combining semantic and lexical matching so agents find exactly what they need.</li><li><strong>Trust:</strong> The search is <strong>access-control aware</strong>. If a human shouldn’t see the data, the agent won’t either.</li></ul><h3>The Bottom Line: Context is the Constraint</h3><p>In the Agentic era, the bottleneck isn’t the model — it’s the context construction. By using Knowledge Catalog as your “Universal Context” engine, you aren’t just storing data; you’re building a dynamic map of your entire business.</p><p>You’re giving your agents a brain, not just a search bar.</p><p><strong>Next up in this series:</strong> We’ll look at <strong>Agentic Capabilities</strong> — how we turn this universal context into autonomous action using the new <strong>Data Agent Kit</strong>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ca4b57dbab77" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond Grounding: Meet the Agentic Data Steward]]></title>
            <link>https://medium.com/google-cloud/beyond-grounding-meet-the-agentic-data-steward-c673d8cef57d?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/c673d8cef57d</guid>
            <category><![CDATA[agents]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Thu, 23 Apr 2026 14:12:21 GMT</pubDate>
            <atom:updated>2026-04-23T14:12:21.941Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qsv2WavUUNcXfGn9fTuoCQ.png" /></figure><p>In my <a href="https://medium.com/google-cloud/teaching-agents-to-read-your-data-a-practical-guide-to-grounding-with-the-dataplex-knowledge-394cb827f931">previous post</a>, we talked about giving your AI a <strong>map</strong> using the Knowledge Catalog. We taught agents how to <em>read</em> your data so they could write better SQL.</p><p>But once an agent can read the map, the next question is: <strong>Who is keeping the map up to date?</strong></p><p>In most organizations, the <strong>“map”</strong> (metadata, glossaries, and policy tags) is a mess. Descriptions are missing, policy tags are inconsistent, and the Data Steward is drowning in manual tickets. I’ve built a prototype to solve this: <a href="https://github.com/akanksha86/governance-metadata-propagation"><strong>The Agentic Data Steward</strong></a><strong>.</strong> It moves beyond just answering queries to actively maintaining the governance health of your entire data warehouse.</p><h3>The Governance Gap: Why “Reading” Isn’t Enough</h3><p>Even with a great grounding context, an agent is only as good as the metadata provided. If your orders table has no description and your user_id column isn’t tagged as PII, the agent—and your users—are flying blind.</p><p>Traditional governance fails because it’s:</p><ul><li><strong>Manual:</strong> A steward must manually type out 500 column descriptions.</li><li><strong>Static:</strong> Metadata doesn’t “flow” when you create a new view or table.</li><li><strong>Silent:</strong> Data quality issues happen in a vacuum, isolated from the business glossary.</li></ul><h3>Enter the Agentic Steward: A Force Multiplier</h3><p>The <strong>Agentic Steward</strong> prototype is designed to act as a digital teammate for Data Stewards. It uses Gemini and BigQuery’s column-level lineage to automate the “boring” parts of governance.</p><p>Here is how it changes the game:</p><h4>1. Automated Metadata Propagation (The “Inheritance” Rule)</h4><p>Why describe customer_id ten times across ten different tables? The Agentic Steward looks at <strong>Lineage</strong>. If the source table is documented either by Steward or using <a href="https://docs.cloud.google.com/bigquery/docs/data-insights">BigQuery Data Insights</a>, the agent automatically propagates those descriptions and glossary terms to downstream derived tables.</p><h4>2. Intelligence-Driven Policy Tagging</h4><p>Assigning policy tags is high-stakes work. One missed column is a compliance risk. The prototype uses LLM-based reasoning to scan schemas and suggest tags. If a column looks like an email or a credit card number, the Steward gets a proactive suggestion to tag it, ensuring the “map” stays secure.</p><h4>3. Governance Health Scoring</h4><p>Instead of wondering “How documented is my dataset?”, the Agentic Steward provides a <strong>Health Check</strong>. It audits your Dataplex entry groups and identifies:</p><ul><li><strong>Missing documentation</strong> and orphaned assets.</li><li><strong>Glossary gaps</strong>, recommending mappings to physical assets.</li><li><strong>Data Quality tracking</strong>, integrating scorecard results directly into the steward’s view.</li></ul><h3>How it Works: The Blueprint</h3><p>The prototype isn’t a “product” — it’s a <strong>blueprint</strong> for what’s possible. It leverages three key components:</p><ul><li><strong>BigQuery &amp; Knowledge Catalog:</strong> The source of truth for schema, lineage, and metadata.</li><li><strong>Gemini Pro:</strong> The reasoning engine that suggests descriptions and tags based on context.</li><li><strong>Agentic Framework:</strong> A loop that checks for “Governance Debt” and proposes fixes for the steward to approve.</li></ul><h3>See it in Action</h3><p>I’m sharing this prototype ahead of our <strong>Google Cloud NEXT</strong> session, where we’ll see how real-world customers are tackling these exact challenges.</p><blockquote><strong><em>Important Note:</em></strong><em> This is an open-source prototype for experimentation and is </em><strong><em>not an official GCP product</em></strong><em>. It is meant to show you how to build your own “governance engine” and to fully use this, you would require column level lineage API allowlisted.</em></blockquote><p><strong>Explore the GitHub Repo here:</strong> <a href="https://github.com/akanksha86/governance-metadata-propagation">akanksha86/governance-metadata-propagation</a></p><p>Stop policing your data and start orchestrating it. <strong>Give your Data Stewards an agent, and let the map start drawing itself.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c673d8cef57d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/google-cloud/beyond-grounding-meet-the-agentic-data-steward-c673d8cef57d">Beyond Grounding: Meet the Agentic Data Steward</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[Teaching Agents to read your Data: A Practical Guide to Grounding with the Dataplex Knowledge…]]></title>
            <link>https://medium.com/google-cloud/teaching-agents-to-read-your-data-a-practical-guide-to-grounding-with-the-dataplex-knowledge-394cb827f931?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/394cb827f931</guid>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[data-governance]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[dataplex]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Fri, 17 Oct 2025 09:36:23 GMT</pubDate>
            <atom:updated>2025-10-24T05:14:00.015Z</atom:updated>
            <content:encoded><![CDATA[<h3>Teaching Agents to read your Data: A Practical Guide to Grounding with the Dataplex Knowledge Engine</h3><p><strong>Authors — Akanksha Bhagwanani and Sai Charan Tej Kommuri</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ifMwyObkfteLa_g3XUpKVw.png" /></figure><p>Imagine asking your data agent,</p><p><strong>“Show me the average order value by region for the last quarter.”</strong></p><p>and getting back… a SQL query that joins the wrong tables, sums the wrong field, or uses <strong><em>state_name</em></strong> instead of <strong><em>country</em></strong>.</p><p>It’s not the model’s fault — it simply doesn’t <strong><em>know your data landscape</em></strong>.<br>It’s like asking a tourist for directions in a city without giving them a map.</p><p>This post is about giving your agent that map.</p><p>It’s about <strong>grounding</strong> — teaching your AI assistant how your data is structured, what each table means, and how things connect. And we’ll use the <strong>Dataplex</strong> <strong>Knowledge Engine</strong> to do it.</p><h3>Why do Agents Struggle with Organization’s Data?</h3><p>Most data warehouses, including BigQuery, have rich metadata hidden in their schema — but agents only see text prompts.</p><p>So when you ask:</p><p>“What’s the repeat purchase rate among loyal customers?”</p><p>the model needs to figure out:</p><ul><li>Which table has orders?</li><li>What column tracks repeat behavior?</li><li>How is “loyal” defined?</li></ul><p>Without context, it guesses. With context, it reasons.</p><h3>The Knowledge Engine JSON: A Map for Machines</h3><p>The <strong>Dataplex Knowledge Engine</strong> converts your dataset metadata into a JSON artifact that an agent can easily consume.</p><p>It’s like a <em>schema memory capsule</em>: a compact, machine-readable document that describes your data universe — tables, columns, relationships, and even example queries.</p><p>Here’s a simplified snippet from <strong>thelook_ecommerce</strong>: [BigQuery Public Dataset]</p><p>{</p><p>“dataset”: “thelook_ecommerce”,</p><p>“tables”: [</p><p>{</p><p>“name”: “orders”,</p><p>“description”: “Customer orders including total cost, status, and timestamps”</p><p>},</p><p>{</p><p>“name”: “users”,</p><p>“description”: “Registered customers with demographic attributes”</p><p>}</p><p>],</p><p>“columns”: [</p><p>{“table”: “orders”, “name”: “user_id”, “dtype”: “INTEGER”, “description”: “Foreign key to users table”},</p><p>{“table”: “orders”, “name”: “total_cost”, “dtype”: “FLOAT”, “description”: “Order value in USD”},</p><p>{“table”: “users”, “name”: “country”, “dtype”: “STRING”, “description”: “Country of residence”}</p><p>],</p><p>“relationships”: [</p><p>{“src_table”: “orders”, “src_column”: “user_id”, “dst_table”: “users”, “dst_column”: “id”, “relation_type”: “fk”}</p><p>],</p><p>“recommended_queries”: [</p><p>{“intent”: “average order value by country”, “sql_template”: “SELECT u.country, AVG(o.total_cost) FROM orders o JOIN users u ON o.user_id = u.id GROUP BY u.country”}</p><p>]</p><p>}</p><p>This JSON becomes the <em>grounding context</em> that you can pass to any LLM-based data agent.</p><h3>How Agents use this Context</h3><p>Let’s walk through the major use cases — with examples from thelook_ecommerce.</p><h3>(a) NL2SQL that actually understands your Schema</h3><p>Without grounding:</p><p>“Total revenue by gender last month”</p><p>→ The model might guess column names like <em>revenue</em>, <em>sales</em>, or <em>gender_code</em> — which may not exist.</p><p>With grounding, the agent reads that:</p><ul><li>orders.total_cost stores revenue</li><li>users.gender stores gender</li><li>orders.created_at stores timestamps</li><li>There’s a user_id → users.id relationship</li></ul><p>So the agent confidently generates:</p><p>SELECT u.gender, SUM(o.total_cost) AS total_revenue</p><p>FROM orders o</p><p>JOIN users u ON o.user_id = u.id</p><p>WHERE o.created_at BETWEEN ‘2025–09–01’ AND ‘2025–09–30’</p><p>GROUP BY u.gender;</p><p>It’s not just syntax; it’s <em>semantic reasoning</em>.</p><h3>(b) Detecting Duplicate Tables and Columns</h3><p>In large enterprises, duplicate tables are a silent cost sink and discovery swamp.</p><p>Suppose your data warehouse has:</p><ul><li>orders, orders_backup, and orders_cleaned_2023<br> All with similar columns like <em>total_cost</em> and <em>user_id</em>.</li></ul><p>The Knowledge Engine JSON captures column names, types, descriptions and relationships, letting an agent spot overlaps:</p><p>“These tables share 92% schema similarity — consider merging or retiring one.”</p><p>This can be extended across projects, too — comparing e-commerce and marketing datasets to identify redundant customer dimension tables.</p><h3>(c) Recommending Glossary Terms</h3><p>Data literacy depends on consistency.</p><p>Let’s say your column is named net_payment_amount but your glossary defines “Revenue” as <em>“Sum of order total excluding refunds and discounts.”</em></p><p>The agent, reading column descriptions and examples, suggests:</p><p>“Column orders.total_cost maps closely to the glossary term Revenue.”</p><p>Over time, your glossary becomes self-updating — not because someone curated it manually, but because your Knowledge Engine gave the agent the clues.</p><h3>(d) Identifying Possible Data Products</h3><p>Sometimes, your datasets already contain the ingredients for data products — you just haven’t named them yet.</p><p>Example: in thelook_ecommerce, frequent joins between:</p><ul><li>orders,</li><li>users, and</li><li>inventory_items</li></ul><p>may indicate a “Customer Purchase Insights” data product.</p><p>The agent, observing these patterns and relationships, could suggest:</p><p>“Propose a data product combining orders, users, and inventory to support revenue analysis by product category.”</p><p>This turns ad hoc analytics into reusable, shareable data assets.</p><h3>(e) Creating a Cross-Dataset Knowledge Graph</h3><p>If you run multiple datasets — say, <em>ecommerce</em>, <em>marketing</em>, and <em>support</em> — their relationships can form a <strong>cross-dataset knowledge graph</strong>.</p><p>For example:</p><ul><li>ecommerce.orders.user_id ↔ marketing.email_campaigns.user_id</li><li>support.tickets.email ↔ users.email</li></ul><p>The agent merges these relationships into a graph persisted in Neo4j, enabling queries like:</p><p>“Show me users who purchased after opening a campaign email.”</p><p>This is the foundation for reasoning over your organization’s entire data landscape.</p><h3>Two ways to Ground your Agents</h3><p>Now that you have the JSON, how do you use it?</p><h3>Option 1: Inline grounding (system instructions)</h3><p>If your agent runs inside a managed environment (like LangChain or Vertex AI Agents), embed the JSON directly in its system prompt:</p><p>You are a data reasoning agent.</p><p>Use the following dataset context to interpret queries:</p><p>{… knowledge_engine.json …}</p><p>The agent will “see” your schema before parsing user questions.</p><h3>Option 2: Retrieval grounding (context-on-demand)</h3><p>If your JSON is large, store it in a vector or graph store (like Pinecone or Neo4j).<br> When a user asks a question, retrieve only the relevant snippet.</p><p>For example:</p><p>User: “Show me total sales by country.”<br> System: Retrieve only nodes related to orders, users, and country.<br> Agent: Generates grounded SQL.</p><p>This keeps grounding lightweight and scalable.</p><h3>How to Generate the JSON Context</h3><p>Simply run the Insights scan on the Dataset of your choice, wait for it to complete and you see the JSON.</p><p>Optional: push it into Neo4j for relationship visualizations and reasoning.</p><h3>A Quick Demonstration</h3><p>Let’s try a simple before–after.</p><p><strong>Ungrounded prompt:</strong></p><p>“Top 5 product categories by revenue.”</p><p>Model guesses incorrectly:</p><p>SELECT category, SUM(revenue) FROM sales GROUP BY category;</p><p>→ sales table doesn’t exist.</p><p><strong>Grounded prompt (with JSON context):</strong></p><p>SELECT p.category, SUM(o.total_cost) AS revenue</p><p>FROM orders o</p><p>JOIN products p ON o.product_id = p.id</p><p>GROUP BY p.category</p><p>ORDER BY revenue DESC</p><p>LIMIT 5;</p><p>Now it’s <em>valid</em>, <em>joins correctly</em>, and <em>uses real column names</em>.<br>That’s the power of grounding.</p><h3>From Query Automation to Data Intelligence</h3><p>Once your agents understand structure, they start understanding meaning.<br>They stop being “SQL translators” and start acting like <em>junior data analysts</em> who can:</p><ul><li>Explain which tables influence a metric,</li><li>Suggest glossary mappings,</li><li>Recommend reusable data products,</li><li>Or even audit your schema for redundancy.</li></ul><p>You don’t just automate data access — you <em>amplify</em> data intelligence.</p><h3>Closing Thought</h3><p>A well-grounded agent is like a data scientist who not only knows SQL syntax but also <em>understands your business model</em>.</p><p>You can hand it any question — “Which products are trending in Germany?” — and trust that it knows which tables to look at and how they connect.</p><p>So before you scale your next data agent, ask yourself:</p><p>“Does it know my data landscape?”</p><p>If not, hand it the map — your <strong>Knowledge Engine JSON</strong> — and watch it finally learn to navigate your data city.</p><p>Also published at — <a href="https://discuss.google.dev/t/teaching-agents-to-read-your-data-a-practical-guide-to-grounding-with-the-dataplex-knowledge-engine/274768">https://discuss.google.dev/t/teaching-agents-to-read-your-data-a-practical-guide-to-grounding-with-the-dataplex-knowledge-engine/274768</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=394cb827f931" width="1" height="1" alt=""><hr><p><a href="https://medium.com/google-cloud/teaching-agents-to-read-your-data-a-practical-guide-to-grounding-with-the-dataplex-knowledge-394cb827f931">Teaching Agents to read your Data: A Practical Guide to Grounding with the Dataplex Knowledge…</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[Data Governance and Sharing Advancements in the Era of Generative AI — Google Cloud Next 2024]]></title>
            <link>https://medium.com/@akankshabhagwanani/data-governance-and-sharing-advancements-in-the-era-of-generative-ai-google-cloud-next-2024-68497810f4e4?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/68497810f4e4</guid>
            <category><![CDATA[data-sharing]]></category>
            <category><![CDATA[dataplex]]></category>
            <category><![CDATA[data-governance]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[bigquery]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Mon, 22 Apr 2024 14:16:57 GMT</pubDate>
            <atom:updated>2024-04-22T14:16:57.118Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Data Governance and Sharing Advancements in the Era of Generative AI</strong> — Google Cloud Next 2024</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*s1-AouB49IRyUgW8WtXXNg.png" /></figure><p>Generative AI is transforming how businesses operate, fueling a greater need for robust data governance. Google Cloud Next 2024 highlights critical advancements in data governance and sharing that enable organizations to harness the power of AI responsibly and effectively. BigQuery, as a unified data platform, places data governance at the forefront of its design. Centralized AI data governance is driven by Dataplex, while seamless data sharing is facilitated by Analytics Hub.</p><h4><strong>Key Advancements in Data Governance and Sharing</strong></h4><ul><li><strong>Enhanced Automatic Cataloging:</strong> Metadata harvesting is now expanded to include BigTable, Spanner, Vertex AI models, and Datasets (Generally Available), with future support for Vertex AI feature store, Cloud SQL, and Looker (Preview). This unified cataloging creates a strong foundation for data-to-AI governance, simplifying metadata management, lineage tracking, and more.</li><li><strong>Data Lineage Advancements:</strong> Lineage support expands to Vertex AI pipelines (Preview), providing complete visibility from data processing to model deployment. Dataplex and Open Lineage API integrations foster third-party extensibility. Column-level lineage (Preview) in BigQuery offers granular impact analysis. Usability enhancements include single-click lineage exploration, filtering, and export options.</li><li><strong>Generative AI-powered Semantic Search (announced):</strong> Natural language queries streamline data discovery, eliminating complex syntax. Semantic search leverages both technical and user-generated metadata for increased precision.</li><li><strong>Data Insights (preview):</strong> Automated query generation and availability tackles the complexities of unfamiliar datasets, providing a faster entry point for exploration. Data insights draw from profile scan data and user queries to offer meaningful starting points.</li><li><strong>Governance Rules:</strong> Metadata is transformed into actionable governance policies through a centralized management interface. Intuitive queries and specifications define precise rules across distributed data, aligning with organizational objectives. Initial focus is on cross-system access policies for BigQuery and Google Cloud Storage which is in Preview.</li><li><strong>Data Sharing Updates:</strong> Data Clean Rooms (part of Analytics Hub) reach General Availability, offering privacy-centric analysis environments. Data providers gain usage metrics and can manage subscriptions. Additional Analytics Hub features include support for user-defined functions, time travel, materialized views over linked datasets, and cross-cloud sharing via BigQuery Omni.</li></ul><p><strong>Conclusion</strong></p><p>AI innovation and strong data governance are intrinsically linked. The announcements at Google Cloud Next 2024 empower organizations to leverage the potential of generative AI while maintaining robust data governance practices.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=68497810f4e4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[BigQuery’s Top 5 Advancements at Google Cloud Next 2024: Expanding Data Analytics Capabilities]]></title>
            <link>https://medium.com/@akankshabhagwanani/bigquerys-top-5-advancements-at-google-cloud-next-2024-expanding-data-analytics-capabilities-17f5f4abc2ef?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/17f5f4abc2ef</guid>
            <category><![CDATA[google-cloud]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[bigquery]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Mon, 22 Apr 2024 08:44:28 GMT</pubDate>
            <atom:updated>2024-04-22T08:44:28.518Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*v2L3DQIA1XUztHKsCfLW_A.png" /></figure><p><strong>Google Cloud Next 2024</strong> showcased major strides for BigQuery, solidifying its position as a <strong>powerful unified data platform</strong>. These enhancements streamline data analytics across diverse formats and use cases. While generative AI innovations have been impressive, this article focuses on five key advancements strengthening BigQuery’s versatility.</p><p><strong>1. BigLake Advancements (Preview)</strong></p><ul><li><strong>Fully Managed Apache Iceberg:</strong> Streamlined data lake advancements with DML support and high-throughput streaming.</li><li><strong>Delta Format Support:</strong> Enhanced flexibility and integration capabilities.</li><li><strong>Cloud Spanner Schema Integration:</strong> Seamlessly utilize Cloud Spanner schemas as external BigQuery datasets.</li></ul><p><strong>2. Enhanced Unstructured Data Support (GA)</strong></p><ul><li><strong>BigLake Object Tables:</strong> Unlocks SQL-based analytics for images, audio, documents, and videos.</li><li><strong>Integration with Vision API, Translate etc.:</strong> Enables rich insights derived from unstructured data, streamlining processes like image classification and translation and creating data pipelines with combination of structure and unstructured data.</li></ul><p><strong>3. BigQuery Continuous Queries (Preview)</strong></p><ul><li><strong>SQL-Based Streaming Analytics:</strong> Facilitates real-time pipeline development for anomaly detection, sentiment analysis and recommendations.</li><li><strong>Mid-Pipeline Integration (Reverse ETL):</strong> Broadens streaming data’s utility, enabling integration with services like Pub/Sub and BigTable.</li></ul><p><strong>4. Serverless Spark Experience (Preview)</strong></p><ul><li><strong>Seamless BigQuery Studio Integration:</strong> Effortlessly author and execute Spark code without data exports or infrastructure management.</li><li><strong>Spark Stored Procedures:</strong> Simplifies complex operations and workflows.</li><li><strong>Integrated Security:</strong> Leverages BigQuery’s security model for robust Spark usage.</li></ul><p><strong>5. BigQuery Streaming Enhancements (Announced)</strong></p><ul><li><strong>Pub/Sub Cross-Cloud Ingestion:</strong> Bridges AWS Kinesis Data Streams to Pub/Sub for expanded real-time analytics.</li><li><strong>Native Pub/Sub BigQuery Subscriptions (Preview):</strong> Streamlined ingestion into BigLake Managed Tables.</li><li><strong>Fully Managed Apache Kafka for BigQuery:</strong> Automated lifecycle management and seamless integration with existing Kafka applications.</li></ul><p><strong>Additional Notable Announcements (GA)</strong></p><ul><li><strong>BQML Advancements:</strong> Enhancements in model development, feature engineering, and explainability.</li><li><strong>BigQuery Omni:</strong> Serverless BigQuery on AWS and Azure for powerful cross-cloud analytics and unified governance.</li></ul><p><strong>Key Takeaways</strong></p><p>These advancements expand BigQuery’s ability to handle structured, semi-structured, and unstructured data, empowering businesses to derive insights from diverse sources in real-time. Its integration with other GCP services and external data platforms solidifies BigQuery’s role as a central hub in modern data analytics architectures.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=17f5f4abc2ef" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[BigQuery’s Top 5 Generative AI Advancements from Google Cloud Next 2024]]></title>
            <link>https://medium.com/@akankshabhagwanani/bigquerys-top-5-generative-ai-advancements-from-google-cloud-next-2024-610bfcf982c0?source=rss-a93648d4a988------2</link>
            <guid isPermaLink="false">https://medium.com/p/610bfcf982c0</guid>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[generative-ai-solution]]></category>
            <category><![CDATA[bigquery]]></category>
            <dc:creator><![CDATA[Akankshabhagwanani]]></dc:creator>
            <pubDate>Tue, 16 Apr 2024 19:54:31 GMT</pubDate>
            <atom:updated>2024-04-16T19:54:31.350Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QIKBuTMX993C25bKf3gwww.png" /></figure><p>Generative AI continues to transform the tech landscape and Google Cloud Next 2024 underscored its integral role in modern data analytics. BigQuery, in particular, unveiled remarkable advancements, placing AI at the forefront of data-driven insights. Let’s delve into the top 5 highlights (all currently in preview):</p><p>First and foremost BigQuery Studio is now in GA which I am not counting towards the top 5 considering it was already in preview announced at Next last year.</p><h4>1. Multimodal Embeddings</h4><p>Embeddings are the building blocks of many ML applications such as semantic search, recommendations, clustering, outlier detection, named entity extraction, and more. Building upon existing text embedding and vector search capabilities, BigQuery now supports multimodal embeddings. This unifies the representation of text and image data within the same semantic space and the generated embeddings have the same dimensionality. This enables a rich array of use cases such as embedding and indexing your images and then searching for them via text.</p><p>Step wise process and details are available in <a href="https://cloud.google.com/blog/products/data-analytics/bigquery-multimodal-embeddings-generation">official blog</a>.</p><h4>2. Data Canvas</h4><p>Data Canvas re-envisions the data-to-insights workflow with an AI-centric approach. Data Canvas helps users create graphical data workflows that map to their mental model supported by AI. Key features include:</p><ul><li><strong>Natural Language Search:</strong> Powered by Dataplex, easily discover data assets using everyday language.</li><li><strong>AI-Assisted SQL Generation:</strong> Create editable SQL queries from natural language input.</li><li><strong>Visualizations and Insights:</strong> Visualize results, glean automated AI-derived insights and seamlessly export to Looker Studio.</li><li><strong>Collaboration:</strong> Share canvases and streamline analysis with colleagues.</li></ul><h4>3. Data Preparation (coming soon)</h4><p>Data Preparation promises an AI-powered experience to simplify the often laborious initial step of data transformation. Expect these features:</p><ul><li><strong>Smart Suggestions:</strong> AI intelligently aids in schema mapping, normalization, enrichment, and quality assurance.</li><li><strong>Visual Wrangling:</strong> Streamlined user experience for intuitive data manipulation.</li><li><strong>CI/CD &amp; Version Control:</strong> Enhancements for collaboration and change management.</li></ul><h4>4. Data Insights</h4><p>Data analysts face the cold-start problem in data exploration, when exploring a new dataset with little or no prior knowledge. The problem often involves uncertainties about the data structure, key patterns, and relevant insights.</p><p>Data Insights tackles the challenges of exploring unfamiliar datasets by intelligently automating query generation. The insights provides valuable cues to help you make informed decisions and gain deeper insights into your data. Rather than starting with a blank slate, you can more quickly start data exploration with meaningful queries that offer valuable insights. Queries generated using data insights are grounded using published <a href="https://cloud.google.com/dataplex/docs/use-data-profiling">profile scan data</a> and queries run by other users of this dataset.</p><h4>5. Gemini in BigQuery</h4><p>Gemini acts as your ever-present AI pair programmer within BigQuery. It facilitates:</p><ul><li><strong>SQL Explanation:</strong> Gain clarity on complex queries, especially those not authored by you.</li><li><strong>SQL &amp; Python Generation and Completion:</strong> Accelerate development with AI-assisted code generation and completion.</li></ul><p>These advancements demonstrate the powerful combination of generative AI and BigQuery. I’ll be following up with additional exciting announcements in the areas of BigQuery (not Generative AI) and data governance.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=610bfcf982c0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>