<?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 Nikhil Kumar Nayak on Medium]]></title>
        <description><![CDATA[Stories by Nikhil Kumar Nayak on Medium]]></description>
        <link>https://medium.com/@nikhilkumarnayak?source=rss-2a7b979d962f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ulLcPubyiwydzkANoMazxw.jpeg</url>
            <title>Stories by Nikhil Kumar Nayak on Medium</title>
            <link>https://medium.com/@nikhilkumarnayak?source=rss-2a7b979d962f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 20:59:52 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nikhilkumarnayak/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[Beyond Chatbots: Building Always-On, Governed Enterprise AI with Databricks Genie, Microsoft…]]></title>
            <link>https://nikhilkumarnayak.medium.com/beyond-chatbots-building-always-on-governed-enterprise-ai-with-databricks-genie-microsoft-570691cc7448?source=rss-2a7b979d962f------2</link>
            <guid isPermaLink="false">https://medium.com/p/570691cc7448</guid>
            <category><![CDATA[databricks-genie]]></category>
            <category><![CDATA[genie]]></category>
            <category><![CDATA[mcp-server]]></category>
            <category><![CDATA[microsoft-foundry]]></category>
            <category><![CDATA[enterprise-ai]]></category>
            <dc:creator><![CDATA[Nikhil Kumar Nayak]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 14:13:02 GMT</pubDate>
            <atom:updated>2026-11-16T14:13:02.711Z</atom:updated>
            <content:encoded><![CDATA[<h3>Beyond Chatbots: Building Always-On, Governed Enterprise AI with Databricks Genie, Microsoft Foundry and MCP🚀</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gJccWFe0Q00ISAD8EWXbcw.png" /></figure><p>Enterprise AI is entering an important new phase.</p><p>The first generation of enterprise generative AI was largely conversational:</p><p><strong>Ask → Retrieve → Generate → Answer</strong></p><p>That model is extremely useful, but much of enterprise work does not begin and end with a single question.</p><p>A claims team continuously monitors cases.</p><p>A finance team continuously watches budgets and forecasts.</p><p>A security team continuously monitors incidents.</p><p>An operations team follows issues until they are resolved.</p><p>A marketing team monitors campaigns and adjusts when performance changes.</p><p>Real enterprise work is often <strong>continuous, event-driven, permission-sensitive and action-oriented</strong>.</p><p>That is why the next evolution of enterprise AI is not simply a better chatbot.</p><p>It is an <strong>always-on, governed AI coworker</strong>.</p><p>Databricks’ new Genie direction provides a useful example of this evolution, while the integration of Databricks Genie with Microsoft Foundry through Model Context Protocol (MCP) demonstrates another critical part of the architecture: how autonomous agents can access enterprise information without abandoning enterprise governance.</p><h3>1. From AI Assistant to AI Coworker</h3><p>A traditional AI assistant normally waits for us.</p><pre>                USER<br>                  │<br>                  ▼<br>                PROMPT<br>                  │<br>                  ▼<br>                AI ASSISTANT<br>                  │<br>                  ▼<br>                RETRIEVE / ANALYZE<br>                  │<br>                  ▼<br>                ANSWER</pre><p>Once the answer is generated, the interaction is effectively finished.</p><p>An autonomous coworker operates differently.</p><pre>                 ┌─────────────────┐<br>                 │ Business Goal   │<br>                 └────────┬────────┘<br>                          ▼<br>                 ┌─────────────────┐<br>                 │ Schedule/Event  │<br>                 │    Trigger      │<br>                 └────────┬────────┘<br>                          ▼<br>                 ┌─────────────────┐<br>                 │ Observe Changes │<br>                 └────────┬────────┘<br>                          ▼<br>                 ┌─────────────────┐<br>                 │ Reason + Search │<br>                 └────────┬────────┘<br>                          ▼<br>                 ┌─────────────────┐<br>                 │ Decide Next     │<br>                 │     Action      │<br>                 └────────┬────────┘<br>                          ▼<br>                    Approval?<br>                    /       \<br>                  YES       NO<br>                   │         │<br>                   ▼         │<br>             Human Review    │<br>                   │         │<br>                   └────┬────┘<br>                        ▼<br>                 ┌───────────────┐<br>                 │ Execute Tool  │<br>                 └───────┬───────┘<br>                         ▼<br>                 ┌───────────────┐<br>                 │Track Progress │<br>                 └───────┬───────┘<br>                         │<br>                         └──────────────► Repeat</pre><p>This is fundamentally different.</p><p>Instead of asking:</p><blockquote><em>“What happened to our claims portfolio last week?”</em></blockquote><p>we can eventually delegate a responsibility such as:</p><blockquote><em>“Continuously monitor incoming claims. Investigate emerging fraud patterns or loss-ratio drift. Alert the appropriate team when thresholds are crossed and prepare the supporting information for review.”</em></blockquote><p>The AI now owns part of a workflow rather than merely answering a question.</p><h3>2. What Makes Databricks Genie “Always-On”?</h3><p>The new Genie is designed around long-running enterprise work.</p><p>Conceptually, its operating loop can be represented as:</p><p><strong>Observe → Understand → Analyze → Decide → Act → Monitor → Repeat</strong></p><p>A job can run on a schedule or respond to a trigger.</p><p>For example:</p><pre>New Claims<br>    │<br>    ▼<br>Monitor continuously<br>    │<br>    ▼<br>Detect unusual pattern<br>    │<br>    ▼<br>Investigate relevant data<br>    │<br>    ▼<br>Fraud / loss-ratio threshold crossed?<br>    │<br>    ├── NO ──► Continue monitoring<br>    │<br>    └── YES<br>          │<br>          ▼<br>     Build context<br>          │<br>          ▼<br>     Recommend action<br>          │<br>          ▼<br>     Human approval<br>          │<br>          ▼<br>     Route / update / notify</pre><p>The crucial word here is not simply <strong>autonomous</strong>.</p><p>It is <strong>governed autonomy</strong>.</p><p>The organization still determines what Genie can access, which tools it can use, which actions can happen automatically, and which require human approval.</p><h3>3. Autonomy Requires Business Context</h3><p>A generic LLM understands language.</p><p>It does not automatically understand what <em>your organization</em> means by:</p><ul><li>Active customer</li><li>Qualified opportunity</li><li>Loss ratio</li><li>Revenue</li><li>High-risk claim</li><li>Approved supplier</li><li>Priority customer</li><li>Critical incident</li></ul><p>Two departments may even use the same term differently.</p><p>This is where <strong>Genie Ontology</strong> becomes important.</p><p>Conceptually:</p><pre>              GENIE ONTOLOGY<br>                    │<br>       ┌────────────┼────────────┐<br>       ▼            ▼            ▼<br>   Definitions    Metrics    Relationships<br>       │            │            │<br>       └────────────┼────────────┘<br>                    ▼<br>            Business Context<br>                    │<br>                    ▼<br>                  Genie</pre><p>The ontology creates a governed semantic understanding of the business across data, definitions, relationships and organizational knowledge.</p><p>Instead of forcing every AI application to independently rediscover what “revenue” means, enterprise agents can operate from shared governed context.</p><h3>4. Finding Information: Genie Search</h3><p>Understanding business terminology solves only half of the problem.</p><p>Enterprise information is distributed across many systems:</p><pre>Databricks Tables ───────┐<br>Dashboards ──────────────┤<br>Documents ───────────────┤<br>SharePoint ──────────────┤<br>Jira ────────────────────┤<br>Confluence ──────────────┤──► Genie Search ──► AI<br>Google Drive ────────────┤<br>Workday ─────────────────┤<br>MCP-connected systems ───┘</pre><p>Genie Search is designed to combine enterprise search with the governed context available to Genie.</p><p>This means an enterprise agent can potentially reason across structured data, documents and connected systems rather than treating each source as an isolated information silo.</p><p>Critically, access control must travel with that information.</p><p>Finding a document does <strong>not</strong> mean every user should be allowed to read it.</p><h3>5. The Security Problem with Enterprise Agents</h3><p>This becomes one of the most important architecture questions in Agentic AI.</p><p>Suppose a user asks a Microsoft Foundry agent:</p><blockquote><em>“What are the highest-risk claims this quarter?”</em></blockquote><p>A simplistic architecture might copy enterprise data into another store or provide the agent with a shared, long-lived credential.</p><p>Both approaches can weaken the existing governance model.</p><p>A stronger principle is:</p><blockquote><strong><em>Bring the agent securely to the governed data instead of unnecessarily bringing governed data to the agent.</em></strong></blockquote><p>The architecture becomes:</p><pre>┌─────────────┐<br>│    USER     │<br>└──────┬──────┘<br>       │<br>       │ User request<br>       ▼<br>┌─────────────────────┐<br>│ Microsoft Foundry   │<br>│       Agent         │<br>└──────────┬──────────┘<br>           │<br>           ▼<br>┌─────────────────────┐<br>│  Foundry Toolbox    │<br>│ Reusable tool layer │<br>└──────────┬──────────┘<br>           │<br>           │ MCP<br>           ▼<br>┌─────────────────────┐<br>│ Per-user OAuth /    │<br>│ Identity Context    │<br>└──────────┬──────────┘<br>           │<br>           ▼<br>┌─────────────────────┐<br>│ Databricks Genie    │<br>└──────────┬──────────┘<br>           │<br>           ▼<br>┌────────────────────────────┐<br>│ Unity Catalog / Gateway    │<br>│ Permissions + Governance   │<br>└────────────┬───────────────┘<br>             │<br>             ▼<br>       Governed Data</pre><p>The important idea is:</p><p><strong>Identity travels to the data.</strong></p><p>The enterprise does not have to give every AI application unrestricted access to a duplicate copy of sensitive data.</p><h3>6. Why MCP Matters</h3><p>Model Context Protocol provides a standardized way for agents to interact with external tools and systems.</p><p>Think of MCP as a standardized tool interface.</p><p>Without such a layer:</p><pre>Agent ──Custom Integration── Databricks<br>Agent ──Custom Integration── ServiceNow<br>Agent ──Custom Integration── GitHub<br>Agent ──Custom Integration── CRM</pre><p>Every integration becomes its own implementation.</p><p>With MCP:</p><pre>┌── Databricks<br>                 │<br>Agent ── MCP ────┼── GitHub<br>                 │<br>                 ├── ServiceNow<br>                 │<br>                 └── Enterprise Tools</pre><p>That does not eliminate security or application-specific logic, but it creates a common protocol for discovering and invoking tools.</p><p>For Databricks, Genie can be exposed to compatible agents through MCP.</p><h3>7. Genie One vs. a Curated Genie Agent</h3><p>There are two useful patterns to understand.</p><h3>Genie One</h3><p>Think of Genie One as the broader enterprise analytical experience.</p><p>It is useful when a user wants business questions answered using governed context across the workspace.</p><pre>Foundry Agent<br>      │<br>      ▼<br> Genie One MCP<br>      │<br>      ▼<br> Genie Ontology<br>      │<br>      ▼<br>Broad governed business context</pre><h3>Curated Genie Agent</h3><p>Sometimes an organization does not want a broad analytical scope.</p><p>A particular team might need a specialized agent containing selected tables, instructions and examples.</p><pre>Foundry Agent<br>      │<br>      ▼<br>Curated Genie Agent<br>      │<br> ┌────┼────┐<br> ▼    ▼    ▼<br>Table Table Metric<br>  A     B    View<br>      +<br> Instructions<br>      +<br>  Examples</pre><p>This provides a narrower domain boundary.</p><p>For example:</p><p><strong>Claims Genie Agent</strong></p><p>could contain only:</p><ul><li>Claim tables</li><li>Policy information</li><li>Loss metrics</li><li>Claims terminology</li><li>Team-specific instructions</li><li>Validated examples</li></ul><p>The correct choice therefore depends on the required scope.</p><h3>8. Why Foundry Toolbox Is Interesting</h3><p>As organizations build more agents, another problem appears.</p><p>Imagine 30 agents independently implementing authentication, MCP configuration and tool integration.</p><p>That becomes difficult to maintain.</p><p>A <strong>Foundry Toolbox</strong> provides a reusable tool layer.</p><pre>Agent A ──────┐<br> Agent B ──────┤<br> Agent C ──────┤<br> Agent D ──────┤<br>               ▼<br>        FOUNDRY TOOLBOX<br>        ┌─────────────┐<br>        │ Genie MCP   │<br>        │ Search      │<br>        │ APIs        │<br>        │ Other Tools │<br>        └──────┬──────┘<br>               ▼<br>       Enterprise Systems</pre><p>Authentication is associated with the connection rather than being repeatedly implemented inside each agent.</p><p>This creates an important separation:</p><p><strong>Agent → Toolbox identity</strong></p><p>and</p><p><strong>Toolbox → downstream data identity</strong></p><p>For per-user scenarios, the downstream call can represent the signed-in user, allowing the target system to enforce that user’s permissions.</p><h3>9. OAuth and Per-User Authorization</h3><p>Consider two employees.</p><pre>Alice<br> │<br> └── Can access Claims A, B, C</pre><pre>Bob<br> │<br> └── Can access Claims D, E</pre><p>They use the same AI agent.</p><p>The dangerous architecture is:</p><pre>Alice ─┐<br>       ├── Agent ── Shared Admin Token ──► ALL DATA<br>Bob ───┘</pre><p>The preferred pattern is:</p><pre>Alice ──► Agent ──► Alice&#39;s delegated identity ──► Allowed data</pre><pre>Bob   ──► Agent ──► Bob&#39;s delegated identity   ──► Allowed data</pre><p>The agent does not become a shortcut around authorization.</p><p>The downstream system still determines what each user can access.</p><p>Foundry can centralize OAuth/token handling through its connections and Toolbox architecture instead of forcing developers to implement token acquisition, storage and refresh inside every agent.</p><h3>10. The First-Time Consent Experience</h3><p>One operational detail is particularly important.</p><p>The first time a user invokes an OAuth-protected MCP tool, consent may be required.</p><p>Conceptually:</p><pre>User asks question<br>       │<br>       ▼<br>Agent selects Genie tool<br>       │<br>       ▼<br>OAuth authorization available?<br>       │<br>     NO│<br>       ▼<br>Consent required<br>       │<br>       ▼<br>User signs in / approves<br>       │<br>       ▼<br>Tool call retried<br>       │<br>       ▼<br>Genie executes query<br>       │<br>       ▼<br>Grounded response</pre><p>This is not necessarily an application failure.</p><p>It can be an expected part of establishing delegated authorization.</p><p>Testing enterprise agents should therefore validate the <strong>real tool execution path</strong>, not simply whether the underlying LLM can independently produce a plausible answer.</p><h3>11. Unity Catalog: Data Governance</h3><p>Unity Catalog provides the data-governance foundation.</p><p>Conceptually:</p><pre>UNITY CATALOG<br>                     │<br>       ┌─────────────┼─────────────┐<br>       ▼             ▼             ▼<br> Permissions       Lineage       Policies<br>       │             │             │<br>       └─────────────┼─────────────┘<br>                     ▼<br>               Governed Data</pre><p>An AI agent should not gain broader data access merely because it is an AI agent.</p><p>Existing authorization remains relevant.</p><p>That is essential for regulated industries such as:</p><ul><li>Healthcare</li><li>Insurance</li><li>Banking</li><li>Life sciences</li><li>Government</li><li>Financial services</li></ul><h3>12. Unity Gateway: Governing AI Actions</h3><p>Data permissions are only one side of autonomous AI.</p><p>Agents also use models and tools.</p><p>That introduces additional governance questions:</p><p><strong>Which MCP servers can this agent access?</strong></p><p><strong>Which tools can it invoke?</strong></p><p><strong>Which actions require approval?</strong></p><p><strong>How much can it spend?</strong></p><p><strong>Can administrators trace what happened?</strong></p><p>Unity Gateway provides a control layer around AI services, MCP tools and agent activity.</p><pre>                  UNITY GATEWAY<br>                       │<br>      ┌────────────────┼────────────────┐<br>      ▼                ▼                ▼<br> Tool Policies     Approvals        Cost Controls<br>      │                │                │<br>      ├────────────────┼────────────────┤<br>      ▼                ▼                ▼<br> Access            Tracing          Monitoring</pre><p>This becomes increasingly important as AI moves from <strong>reading</strong> information to <strong>changing</strong> enterprise systems.</p><h3>13. Human-in-the-Loop Is a Feature, Not a Limitation</h3><p>Autonomous does not have to mean unsupervised.</p><p>Consider:</p><pre>Low-risk operation<br>       │<br>       ▼<br>Execute automatically<br></pre><pre>High-risk operation<br>       │<br>       ▼<br>Generate recommendation<br>       │<br>       ▼<br>Request approval<br>       │<br>       ▼<br>Human reviews<br>       │<br>    ┌──┴──┐<br> Approve Reject<br>    │<br>    ▼<br> Execute</pre><p>Organizations can decide which operations are autonomous and which require sign-off.</p><p>For example:</p><p><strong>Autonomous</strong></p><ul><li>Search information</li><li>Analyze trends</li><li>Monitor metrics</li><li>Prepare summaries</li><li>Generate reports</li></ul><p><strong>Approval required</strong></p><ul><li>Change financial budgets</li><li>Update sensitive records</li><li>Contact customers</li><li>Close incidents</li><li>Execute high-impact business actions</li></ul><p>The objective is therefore not maximum autonomy.</p><p>It is <strong>appropriate autonomy</strong>.</p><h3>14. Putting Everything Together</h3><p>The complete architecture can be visualized as:</p><pre>                  BUSINESS USER<br>                         │<br>                         ▼<br>              MICROSOFT FOUNDRY AGENT<br>                         │<br>                  Reason / Orchestrate<br>                         │<br>                         ▼<br>                  FOUNDRY TOOLBOX<br>                         │<br>                  OAuth + Identity<br>                         │<br>                         ▼<br>                        MCP<br>                         │<br>                         ▼<br>                DATABRICKS GENIE<br>                 /             \<br>                /               \<br>        Genie One          Genie Agent<br>             │                  │<br>             └────────┬─────────┘<br>                      ▼<br>                GENIE ONTOLOGY<br>             Business understanding<br>                      │<br>                      ▼<br>                 GENIE SEARCH<br>            Enterprise discovery<br>                      │<br>          ┌───────────┴───────────┐<br>          ▼                       ▼<br>   UNITY CATALOG            UNITY GATEWAY<br>   Data permissions         Tool governance<br>   Access control           Approval policies<br>   Governance               Tracing / costs<br>          │                       │<br>          └───────────┬───────────┘<br>                      ▼<br>            GOVERNED ENTERPRISE<br>                   DATA<br>                      │<br>                      ▼<br>           Grounded result/action<br>                      │<br>                      ▼<br>               HUMAN APPROVAL<br>                when required<br>                      │<br>                      ▼<br>             CONTINUE MONITORING</pre><p>Now combine that architecture with always-on execution:</p><pre>┌─────────────────────────────┐<br>          │                             │<br>          ▼                             │<br>       Observe                          │<br>          │                             │<br>          ▼                             │<br>       Understand                       │<br>          │                             │<br>          ▼                             │<br>       Retrieve                         │<br>          │                             │<br>          ▼                             │<br>        Reason                          │<br>          │                             │<br>          ▼                             │<br>        Decide                          │<br>          │                             │<br>          ▼                             │<br>   Approval if needed                   │<br>          │                             │<br>          ▼                             │<br>         Act                            │<br>          │                             │<br>          ▼                             │<br>        Track ──────────────────────────┘</pre><p>This is the architecture of an <strong>always-on governed AI coworker</strong>.</p><h3>15. Example: Insurance Claims</h3><p>Consider an insurance organization.</p><p>Instead of an analyst repeatedly asking:</p><blockquote><em>“Are there unusual fraud patterns?”</em></blockquote><p>an ongoing job could monitor claims continuously.</p><pre>Incoming Claims<br>      │<br>      ▼<br>Genie monitors metrics<br>      │<br>      ▼<br>Pattern changes<br>      │<br>      ▼<br>Retrieve supporting claims<br>      │<br>      ▼<br>Apply governed business context<br>      │<br>      ▼<br>Investigate anomaly<br>      │<br>      ▼<br>Prepare case information<br>      │<br>      ▼<br>Alert underwriter<br>      │<br>      ▼<br>Human decision<br>      │<br>      ▼<br>Continue monitoring</pre><p>The value does not come from replacing the underwriter.</p><p>The value comes from removing the repetitive work of continuously searching, monitoring, assembling and preparing information.</p><p>The human remains responsible for decisions where human judgment matters.</p><h3>16. The Bigger Architecture Lesson</h3><p>There are four ideas I believe are particularly important.</p><h3>16.1. Move from prompts to responsibilities</h3><p>The next generation of agents will increasingly own persistent goals instead of isolated conversations.</p><h3>16.2. Keep identity attached to access</h3><p>AI should not become a mechanism for bypassing enterprise authorization.</p><h3>16.3. Keep governance close to the data and tools</h3><p>Data permissions, tool policies, approvals, auditing and cost controls need to remain part of the architecture.</p><h3>16.4. Design autonomy with human checkpoints</h3><p>Enterprise AI should know not only <strong>how to act</strong>, but also <strong>when not to act without approval</strong>.</p><h3>Conclusion</h3><p>The interesting part of the new Databricks Genie story is not simply another AI feature.</p><p>It represents a broader transition:</p><p><strong>Chatbots → Copilots → Agents → Always-on AI coworkers</strong></p><p>And connecting systems such as Microsoft Foundry and Databricks through MCP shows how that transition can happen while preserving enterprise identity and governance.</p><p>The architecture is ultimately about balancing two forces:</p><p><strong>Autonomy</strong></p><p>and</p><p><strong>Control</strong></p><p>Give AI enough context and tools to perform meaningful work.</p><p>But keep identity, authorization, approvals, auditing and governance around every important boundary.</p><p>The future of enterprise AI may therefore not be defined by the assistant that gives the best answer.</p><p>It may be defined by the AI coworker that can <strong>keep working safely after the conversation ends.</strong></p><h3>References</h3><ol><li>Databricks — <em>The new Genie: autonomous AI for complex enterprise work</em></li><li>Databricks / Microsoft Learn — <em>Genie One MCP server</em></li><li>Microsoft Learn — <em>Use Azure Databricks Genie in Microsoft Foundry</em></li><li>Microsoft Learn — <em>Connect agents to MCP server endpoints</em></li><li>Microsoft Learn — <em>Toolbox authentication in Microsoft Foundry</em></li><li>Microsoft Learn — <em>Genie Agent MCP server</em></li><li>Databricks / Microsoft Learn — <em>Unity Gateway documentation</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=570691cc7448" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GenAI: 5-Day Gen AI Intensive Course with Google & Kaggle]]></title>
            <link>https://nikhilkumarnayak.medium.com/genai-5-day-gen-ai-intensive-course-with-google-kaggle-57f7c25538b8?source=rss-2a7b979d962f------2</link>
            <guid isPermaLink="false">https://medium.com/p/57f7c25538b8</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[course]]></category>
            <category><![CDATA[kaggle]]></category>
            <category><![CDATA[google]]></category>
            <dc:creator><![CDATA[Nikhil Kumar Nayak]]></dc:creator>
            <pubDate>Mon, 11 Nov 2024 16:43:31 GMT</pubDate>
            <atom:updated>2024-11-12T16:52:59.213Z</atom:updated>
            <content:encoded><![CDATA[<p>🤩 I came across a free course on Generative AI🌈, taught by the experts at Google and Kaggle! 🤩 Here’s what each day of the course will cover:</p><p>🔌 Day 1: 🤖 Foundational models &amp; prompt engineering 🗣 — from transformers to prompt engineering techniques<br>🌐 Day 2: 🤔 Embeddings &amp; Vector databases 📡 — deep dive into embeddings and vector search applications<br>🔗 Day 3: 🤖 Generative AI Agents 💬 — building sophisticated agents step-by-step<br>🔬 Day 4: 💉 Domain-specific LLMs 🧬 — insights into models like SecLM and Med-PaLM<br>🔧 Day 5: MLOps for GenAI — practical MLOps practices with Vertex AI 🏭</p><p>This is a great opportunity for anyone curious about AI, with no prior knowledge needed! <br>I just have an eagerness to learn, a Kaggle account, access to AI Studio, and lots of curiosity! 😄</p><p>Check it out here for details:</p><h4>What’s being covered?</h4><blockquote><strong>Day 1:</strong> Foundational Models &amp; Prompt Engineering — Explore the evolution of LLMs, from transformers to techniques like fine-tuning and inference acceleration. Get trained with the art of prompt engineering for optimal LLM interaction.<br>Day1 Course Link:- <a href="https://www.youtube.com/watch?v=dnGuDNhD3Ag">https://youtu.be/kpRyiJUUFxY?si=1DMVO9YST6aoHekq</a></blockquote><blockquote><strong>Day 2:</strong> Embeddings and Vector Stores/Databases — Learn about the conceptual underpinning of embeddings and vector databases, including embedding methods, vector search algorithms, and real-world applications with LLMs, as well as their tradeoffs.<br>Day2 Course Link:- <a href="https://www.youtube.com/watch?v=86GZC56rQCc&amp;list=PLqFaTIg4myu-b1PlxitQdY0UYIbys-2es&amp;index=2">https://www.youtube.com/live/86GZC56rQCc?si=1FHiiyTLrVbALPfO</a></blockquote><blockquote><strong>Day 3: </strong>Generative AI Agents — Learn to build sophisticated AI agents by understanding their core components and the iterative development process.<br>Day3 Course Link:- <a href="http://Day3 Course Link">https://www.youtube.com/watch?v=HQUtMWoTAD4&amp;list=PLqFaTIg4myu-b1PlxitQdY0UYIbys-2es&amp;index=3</a></blockquote><blockquote><strong>Day 4:</strong> Domain-Specific LLMs — Delve into the creation and application of specialized LLMs like SecLM and Med-PaLM, with insights from the researchers who built them.<br>Day4 Course Link:- <a href="https://www.youtube.com/watch?v=odvuLMJWUSU&amp;list=PLqFaTIg4myu-b1PlxitQdY0UYIbys-2es&amp;index=4">https://www.youtube.com/watch?v=odvuLMJWUSU&amp;list=PLqFaTIg4myu-b1PlxitQdY0UYIbys-2es&amp;index=4</a></blockquote><blockquote><strong>Day 5:</strong> MLOps for Generative AI — Discover how to adapt MLOps practices for Generative AI and leverage Vertex AI’s tools for foundation models and generative AI applications.<br>Day5 Course Link:- <a href="https://www.youtube.com/watch?v=uCFW0i9xrBc&amp;list=PLqFaTIg4myu-b1PlxitQdY0UYIbys-2es&amp;index=5">https://www.youtube.com/watch?v=uCFW0i9xrBc&amp;list=PLqFaTIg4myu-b1PlxitQdY0UYIbys-2es&amp;index=5</a></blockquote><p>Subscribe to this channel for further learning:<br><em>Kaggle Youtube channel:</em> <a href="https://lnkd.in/gA3Hu4kj">https://lnkd.in/gA3Hu4kj</a></p><p><em>Kaggle Community Guidelines</em>: <a href="https://lnkd.in/gg-g9gsE">https://lnkd.in/gg-g9gsE</a></p><p><em>Discord Group(ask questions)</em>: <strong>#5dgai-q-and-a</strong> <a href="https://lnkd.in/geKYkCKw">https://lnkd.in/geKYkCKw</a></p><p><em>Discord Group(course announcements)</em>: <strong>#5dgai-announcements</strong> <br><a href="https://lnkd.in/gjf9rmVp">https://lnkd.in/gjf9rmVp</a></p><p>Let’s learn together!!! So Please join and learn the <a href="https://www.linkedin.com/feed/hashtag/?keywords=genai&amp;highlightedUpdateUrns=urn%3Ali%3Aactivity%3A7261769535144304640">#GenAI</a> concept.</p><p><em>This program is Host by</em> <a href="https://www.linkedin.com/in/dynamicwebpaige/"><strong>Paige Bailey</strong></a> <br><a href="https://www.linkedin.com/company/kaggle/"><strong>Kaggle</strong></a><strong> </strong><a href="https://www.linkedin.com/company/google/"><strong>Google</strong></a><strong> </strong><a href="https://www.linkedin.com/company/experts-program/"><strong>Google Developer Experts</strong></a><strong> </strong><a href="https://www.linkedin.com/company/discord/"><strong>Discord</strong></a><br><em>Special Guest from Google:</em><br><a href="https://www.linkedin.com/in/aminbarekatain/"><strong>Mohammadamin Barekatain</strong></a><strong> </strong><a href="https://www.linkedin.com/in/leeboonstra/"><strong>Lee Boonstra</strong></a><strong> </strong><a href="https://www.linkedin.com/in/logankilpatrick/"><strong>Logan Kilpatrick</strong></a><strong> </strong><a href="https://www.linkedin.com/in/daniel-j-mankowitz-96a25a46/"><strong>Daniel J. Mankowitz</strong></a><strong> </strong><a href="https://www.linkedin.com/in/majdalmerey/"><strong>Majd Al Merey</strong></a><strong> </strong><a href="https://www.linkedin.com/in/anant-nawalgaria/"><strong>Anant Nawalgaria</strong></a><strong> </strong><a href="https://www.linkedin.com/in/aseveryn/"><strong>Aliaksei Severyn</strong></a><strong> </strong><a href="https://www.linkedin.com/in/chuck-sugnet/"><strong>Chuck Sugnet</strong></a></p><p><strong>#LearnMoreShareMore #AI #ML #MLOps #Kaggle #Google #Discord</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=57f7c25538b8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Top LLMs that are Built in INDIA]]></title>
            <link>https://nikhilkumarnayak.medium.com/top-llms-that-are-built-in-india-edf242648205?source=rss-2a7b979d962f------2</link>
            <guid isPermaLink="false">https://medium.com/p/edf242648205</guid>
            <category><![CDATA[research]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Nikhil Kumar Nayak]]></dc:creator>
            <pubDate>Sat, 01 Jun 2024 12:34:01 GMT</pubDate>
            <atom:updated>2024-06-01T12:34:01.595Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mBOmyK3SdJU9PzXAiRwOrA.png" /></figure><p><strong>LLM </strong>stands for <strong>Large Language Model</strong>. It refers to large neural network models trained on vast amounts of text data to understand and generate human-like text.</p><p>Some of the most well-known LLMs include GPT (Generative Pre-trained Transformer) models like GPT-3 developed by OpenAI, as well as models like BERT (Bidirectional Encoder Representations from Transformers) and LaMDA (Language Model for Dialogue Applications) from Google.</p><p>Similarly, as we all know India is now in AI Leap. These LLMs have been built in India. Those are,</p><ol><li>OpenHathi</li><li>Tamil-LLAMA</li><li>Krutrim</li><li>Project Indus</li><li>Bhashini</li><li>CoRover.ai</li><li>BharatGPT (By CoRover)</li><li>Hanooman (By BharatGPT)</li><li>Odia Llama</li><li>Dhenu 1.0</li><li>Kannada Llama</li><li>Navarasa 2.0</li><li>More to be added…</li></ol><p>These LLMs contribute to the growth of AI and ML in several ways:</p><ul><li>They advance the state-of-the-art in natural language understanding and generation, pushing the boundaries of what is possible with language models.</li><li>They enable the development of more sophisticated and human-like AI systems that can communicate and interact with users in natural language.</li><li>The training and fine-tuning of LLMs provide insights and advancements in areas like transfer learning, few-shot learning, and multi-task learning, which can benefit other areas of AI and ML.</li><li>The success of LLMs has spurred further research and investment in large-scale model development, potentially leading to breakthroughs in other domains as well.</li></ul><p>LLMs are a crucial component of modern AI. They are helping to drive progress in areas like natural language processing, conversational AI, and content generation, enabling more human-like and intelligent systems.</p><p>#AI #ML #Gen-AI #LLM #DataScience #Learning #Sharing</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=edf242648205" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>