<?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 Sangeeth Sivakumar on Medium]]></title>
        <description><![CDATA[Stories by Sangeeth Sivakumar on Medium]]></description>
        <link>https://medium.com/@sangeethksivakumar?source=rss-27480e9e1ebf------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*1W_ruPQtI5zuhOm35FRsKQ.jpeg</url>
            <title>Stories by Sangeeth Sivakumar on Medium</title>
            <link>https://medium.com/@sangeethksivakumar?source=rss-27480e9e1ebf------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 18:26:48 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@sangeethksivakumar/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 $0 Home Server: Hosting a Production-Grade Agentic AI App ]]></title>
            <link>https://medium.com/@sangeethksivakumar/the-0-home-server-hosting-a-production-grade-agentic-ai-app-9d8b1f4632b5?source=rss-27480e9e1ebf------2</link>
            <guid isPermaLink="false">https://medium.com/p/9d8b1f4632b5</guid>
            <category><![CDATA[homelab]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[cloudflare-tunnel]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Sangeeth Sivakumar]]></dc:creator>
            <pubDate>Sun, 05 Apr 2026 16:06:22 GMT</pubDate>
            <atom:updated>2026-04-05T16:06:22.271Z</atom:updated>
            <content:encoded><![CDATA[<p>Have you ever wondered if that old Raspberry Pi or long-forgotten Mac Mini sitting in your drawer could run a fully automated, AI-driven production app?</p><p>Over the last few hours, I turned a piece of “junk” hardware into a high-performance <strong>AI-Agentic Server</strong>. Using nothing but free tools and a spare computer, I built a production-grade application reachable at <a href="https://ottdates.in/"><strong>https://ottdates.in</strong></a> that updates itself, manages its own logic, and costs <strong>$0 in monthly recurring fees.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qBawYWKD88GGSYAtr80fxw.png" /><figcaption><a href="https://ottdates.in/"><strong>https://ottdates.in</strong></a></figcaption></figure><p>This isn’t just about an OTT tracker — it’s about a new blueprint for building globally accessible services from your living room.</p><h3>🏗️ The Problem: The “Cloud Tax”</h3><p>Building a modern app for developers usually means recurring bills: $20/mo for a VPS, $10/mo for a database, and extra for search APIs and AI credits. I wanted to see if I could build a professional-grade VPC (Virtual Private Cloud) using zero-cost infrastructure.</p><h3>🛠️ The “Zero Cost” Blue-Print</h3><p>The goal: <strong>Professional results, Zero overhead.</strong></p><ol><li><strong>The Brain (AI Logic)</strong>: I used <strong>Google Gemini 1.5 Flash</strong> (via free AI Studio) to handle natural language processing and data extraction.</li><li><strong>The Discovery Engine</strong>: <strong>Serper.dev</strong> (Free Tier) to power the web-wide search functionality.</li><li><strong>The Web Stack</strong>: A modern <strong>Next.js</strong> and <strong>Tailwind CSS</strong> application, optimized for sub-second performance.</li><li><strong>The Hardware (Local Node)</strong>: A spare <strong>Raspberry Pi</strong> (but an old <strong>Mac Mini</strong> or even an old laptop works just as well!).</li><li><strong>The Gateway</strong>: <strong>Cloudflare Tunnels</strong> (cloudflared) to bypass my home ISP and map the local server to a custom domain — <a href="https://ottdates.in/"><strong>ottdates.in</strong></a> — without opening any firewall ports.</li><li><strong>The Agentic Designer</strong>: I used <strong>Antigravity</strong>, an AI-Agentic coding assistant, to co-pilot the entire system architecture in a single afternoon.</li></ol><h3>Why I Chose Tunnels Over Port Forwarding</h3><p><strong>Cloudflare Tunnels</strong> create a secure, outbound-only connection that keeps my home IP hidden and my router’s firewall completely closed. Instead of “punching a hole” in my network for the world to see, the tunnel acts as a private bridge that only allows legitimate traffic to reach my app. For a production-grade setup, you could also leverage <strong>Cloudflare WAF, Geoblocking, Zero Trust Access, and Rate Limiting</strong> to filter out malicious bots and unauthorized users before they ever touch my hardware.</p><h3>🧠 The Orchestrator: From Coding to Intent</h3><p>The real shift here isn’t just using AI to write snippets; it’s about <strong>Intent-Driven Architecture</strong>. By using an <strong>Agentic Developer (Antigravity)</strong>, I moved from managing syntax to managing the entire system goal.</p><p>Instead of manual plumbing, I provided a high-level “Master Prompt” to define the autonomous logic:</p><blockquote>“Build a Node.js sync script that searches for Indian OTT releases via Serper, uses Gemini Flash to extract a clean JSON dataset from raw news, and merges it into my local DB. If a poster is missing, procedurally generate a vibrant, aesthetic CSS-neon poster based on the movie’s title hash. Ensure the loop is optimized for a Raspberry Pi.”</blockquote><p>This turned the server into a <strong>Self-Healing system</strong>. The logic doesn’t just “run” — it adapts. If a source site changes its layout or a link breaks, the agentic layer identifies the failure and triggers a stylish CSS fallback automatically. I’m not just hosting an app; I’m hosting a system that maintains itself while I sleep. 🚀</p><h3>🚀 The Automated Deployment</h3><p>Using a <strong>Linux Cron job</strong> and <strong>PM2</strong> (Process Manager), the server runs a weekly maintenance cycle:</p><ul><li>At 6:00 AM every Friday, it hunts for new data.</li><li>It rebuilds the production website natively on the local hardware.</li><li>It hot-reloads the instance with zero downtime.</li></ul><p>To monitor health, I installed <strong>Netdata</strong> on the local node, giving me a professional, real-time dashboard accessible from any device on my network.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YVLJqseG5wE5PDiTqvOeaw.png" /><figcaption>netdata dashboard for pi</figcaption></figure><h3>💡 The Takeaway: Dust Off Your Old Tech</h3><p>The most important lesson? <strong>You don’t need a high-end cloud server to build high-end software.</strong></p><p>Whether you have an old <strong>Raspberry Pi</strong>, an <strong>old Mac Mini</strong>, or a <strong>dusty PC</strong> from 2012, you have the components for a globally accessible, AI-powered VPC.</p><h3>Project Stats:</h3><ul><li><strong>Infrastructure Cost</strong>: $0 Monthly. (Optional few $ for a custom domain, or FREE if you use an existing subdomain).</li><li><strong>Building Time</strong>: From zero to live in ~4 hours including pi setup.</li><li><strong>Performance</strong>: <strong>200ms</strong> FCP (Faster than many commercial sites hosted on expensive cloud nodes).</li></ul><h3>📝 The Final Loop: Agentic Content</h3><p>To bring the project full circle, I didn’t even write this blog post manually. Once the server was live, I asked <strong>Antigravity</strong> to analyze the entire codebase and generate the initial draft for this Medium article. From the system architecture on my <strong>Mac mini</strong> to the words you’re reading now, the entire workflow was driven by the same agentic intent.</p><p><strong>Check it out live:</strong> 👉 <a href="https://ottdates.in/"><strong>https://ottdates.in</strong></a></p><p><strong>Ready to build your first local AI server?</strong> Let me know in the comments. Let’s stop paying for hosting and start building from home!</p><p>#RaspberryPi #MacMini #SelfHosting #Cloudflare #AI #WebDevelopment #Automation #NextJS #ProjectShowcase</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9d8b1f4632b5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[MCP: Beyond the Hype — What It Actually Standardizes (And What It Doesn’t)]]></title>
            <link>https://medium.com/@sangeethksivakumar/mcp-beyond-the-hype-what-it-actually-standardizes-and-what-it-doesnt-9d35f55c50d1?source=rss-27480e9e1ebf------2</link>
            <guid isPermaLink="false">https://medium.com/p/9d35f55c50d1</guid>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[mcp-server]]></category>
            <category><![CDATA[model-context-protocol]]></category>
            <category><![CDATA[aritificial-intelligence]]></category>
            <category><![CDATA[ai-tools]]></category>
            <dc:creator><![CDATA[Sangeeth Sivakumar]]></dc:creator>
            <pubDate>Fri, 05 Dec 2025 03:45:02 GMT</pubDate>
            <atom:updated>2025-12-05T03:48:17.138Z</atom:updated>
            <content:encoded><![CDATA[<h3>MCP: Beyond the Hype — What It Actually Standardizes (And What It Doesn’t)</h3><p><strong>A practical developer’s perspective on Model Context Protocol</strong></p><p><a href="https://sangeeth.page/"><em>https://sangeeth.page/</em></a></p><h3>Introduction</h3><p>As AI development matures, we’re seeing frameworks emerge that promise to standardize how AI agents interact with tools and data. Anthropic’s Model Context Protocol (MCP) has generated significant buzz in the developer community — and has now been adopted by OpenAI, Google, and major platforms like Cursor, Replit, and Sourcegraph.</p><p>But what does MCP actually standardize? What problems remain unsolved? Let’s examine MCP critically, separating the genuine innovation from the areas where developers still need to do the work themselves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YoVa-cnNNvaZ3HHLqz6fRw.jpeg" /></figure><h3>1. The Foundation: Tool Calling is the Real Game Changer</h3><p>Before diving into MCP, it’s crucial to understand that <strong>tool calling itself is the transformative capability</strong>, not any specific protocol around it.</p><h3>What Tool Calling Enables</h3><p>Tool calling allows AI models to break free from their training data constraints. Instead of just generating text, models can now:</p><ul><li>Query real-time data (weather, stock prices, database records)</li><li>Perform computations (calculations, data transformations)</li><li>Take actions (send emails, create files, update databases)</li><li>Access external systems (APIs, databases, filesystems)</li></ul><h3>A Basic Example (Claude API)</h3><pre>const tools = [{<br>  name: &quot;get_weather&quot;,<br>  description: &quot;Get current weather for a city&quot;,<br>  input_schema: {<br>    type: &quot;object&quot;,<br>    properties: {<br>      city: { type: &quot;string&quot; }<br>    },<br>    required: [&quot;city&quot;]<br>  }<br>}];</pre><pre>const response = await anthropic.messages.create({<br>  model: &#39;claude-sonnet-4-5-20250929&#39;,<br>  tools: tools,<br>  messages: [{ role: &#39;user&#39;, content: &#39;What\&#39;s the weather in Tokyo?&#39; }]<br>});</pre><pre>// Claude responds with tool_use request<br>if (response.stop_reason === &#39;tool_use&#39;) {<br>  const toolCall = response.content.find(c =&gt; c.type === &#39;tool_use&#39;);<br>  // Execute: fetch weather API<br>  // Return result to Claude<br>  // Claude formulates natural language response<br>}</pre><p>This capability exists across all major AI providers (OpenAI, Google, Anthropic) — each with different formats, but the same fundamental concept. <strong>This is the innovation that matters.</strong></p><h3>2. What MCP Actually Is</h3><p>MCP introduces a client-server architecture for managing tools. Here’s the architecture:</p><pre>┌──────────────────────────────────┐<br>│       Your Application           │<br>│  ┌────────────────────────────┐  │<br>│  │       MCP Client           │  │<br>│  │  - Discovers tools         │  │<br>│  │  - Calls tools via protocol│  │<br>│  └────────────┬───────────────┘  │<br>└───────────────┼──────────────────┘<br>                │ MCP Protocol<br>┌───────────────▼──────────────────┐<br>│         MCP Server               │<br>│  ┌────────────────────────────┐  │<br>│  │  Exposes:                  │  │<br>│  │   - tools/list endpoint    │  │<br>│  │   - tools/call endpoint    │  │<br>│  │                            │  │<br>│  │  Implements:               │  │<br>│  │   - get_weather()          │  │<br>│  │   - search_database()      │  │<br>│  └────────────────────────────┘  │<br>└──────────────────────────────────┘</pre><h3>Simple MCP Server Example</h3><pre>import { Server } from &quot;@modelcontextprotocol/sdk/server/index.js&quot;;</pre><pre>const server = new Server({ name: &quot;weather-server&quot;, version: &quot;1.0.0&quot; });</pre><pre>// Endpoint 1: List available tools<br>server.setRequestHandler(&quot;tools/list&quot;, async () =&gt; ({<br>  tools: [{<br>    name: &quot;get_weather&quot;,<br>    description: &quot;Get weather for a city&quot;,<br>    input_schema: {<br>      type: &quot;object&quot;,<br>      properties: { city: { type: &quot;string&quot; } }<br>    }<br>  }]<br>}));</pre><pre>// Endpoint 2: Execute tool<br>server.setRequestHandler(&quot;tools/call&quot;, async (request) =&gt; {<br>  if (request.params.name === &quot;get_weather&quot;) {<br>    const weather = await fetchWeatherAPI(request.params.arguments.city);<br>    return { content: [{ type: &quot;text&quot;, text: JSON.stringify(weather) }] };<br>  }<br>});</pre><h3>Key Concepts</h3><ol><li><strong>Dynamic Discovery</strong>: Clients discover tools at runtime via tools/list</li><li><strong>Standardized Execution</strong>: All tools called via tools/call with consistent structure</li><li><strong>Server-Side Logic</strong>: Tool implementations live in the server, not client applications</li><li><strong>Transport Flexibility</strong>: Supports HTTP/SSE for remote servers, stdio for local processes</li></ol><h3>3. Alternative Approach: Your Own Tool Server</h3><p>Here’s where it gets interesting. You can achieve the same result without MCP by building a simple REST API.</p><h3>Spring Boot Implementation</h3><pre>@RestController<br>@RequestMapping(&quot;/tools&quot;)<br>public class ToolsController {<br>    <br>    @PostMapping(&quot;/list&quot;)<br>    public ToolsListResponse listTools() {<br>        return new ToolsListResponse(Arrays.asList(<br>            new Tool(&quot;get_weather&quot;, &quot;Get weather for a city&quot;, weatherSchema),<br>            new Tool(&quot;query_database&quot;, &quot;Query user database&quot;, dbSchema)<br>        ));<br>    }<br>    <br>    @PostMapping(&quot;/call&quot;)<br>    public ToolCallResponse callTool(@RequestBody ToolCallRequest request) {<br>        switch(request.getName()) {<br>            case &quot;get_weather&quot;:<br>                return weatherService.getWeather(request.getArguments().get(&quot;city&quot;));<br>            case &quot;query_database&quot;:<br>                return dbService.query(request.getArguments().get(&quot;query&quot;));<br>            default:<br>                throw new ToolNotFoundException(request.getName());<br>        }<br>    }<br>}</pre><h3>Client Implementation</h3><pre>// Fetch tools from YOUR server<br>const toolsResponse = await fetch(&#39;https://your-api.com/tools/list&#39;, {<br>  method: &#39;POST&#39;,<br>  headers: { &#39;Authorization&#39;: &#39;Bearer YOUR_TOKEN&#39; }<br>});<br>const tools = await toolsResponse.json();</pre><pre>// Use with Claude<br>const response = await anthropic.messages.create({<br>  tools: tools.tools,<br>  messages: conversationHistory<br>});</pre><pre>// Execute tool via YOUR server<br>if (response.stop_reason === &#39;tool_use&#39;) {<br>  const toolCall = response.content.find(c =&gt; c.type === &#39;tool_use&#39;);<br>  const result = await fetch(&#39;https://your-api.com/tools/call&#39;, {<br>    method: &#39;POST&#39;,<br>    headers: { &#39;Authorization&#39;: &#39;Bearer YOUR_TOKEN&#39; },<br>    body: JSON.stringify({<br>      name: toolCall.name,<br>      arguments: toolCall.input<br>    })<br>  });<br>  // Send result back to Claude<br>}</pre><p><strong>This works perfectly!</strong> The differences from MCP are:</p><ul><li>You choose your endpoint names</li><li>You handle authentication your way</li><li>You control the entire stack</li></ul><h3>So What Does MCP Actually Add?</h3><ol><li><strong>Convention</strong>: Agreed-upon endpoint names (tools/list, tools/call)</li><li><strong>SDK Convenience</strong>: Handles transport details (HTTP, stdio, error handling)</li><li><strong>Ecosystem</strong>: Pre-built MCP servers you can use (GitHub, databases, filesystems)</li><li><strong>Industry Adoption</strong>: OpenAI, Google, and major platforms now support it</li></ol><h3>4. The Critical Question: What Does MCP Standardize vs. What It Doesn’t</h3><p>This is where the nuance matters. Let’s be precise.</p><h3>What MCP Standardizes ✓</h3><p>Layer What’s Standardized <strong>Transport Protocol</strong> tools/list, tools/call endpoints <strong>Transport Options</strong> HTTP/SSE, stdio <strong>Server Schema Format</strong> All MCP servers use the same tool definition format <strong>Client-Server Pattern</strong> Architectural approach for tool discovery and execution</p><h3>What MCP Does NOT Standardize ✗</h3><p>Layer What’s NOT Standardized <strong>LLM Input Format</strong> Each model expects tools in its own format <strong>Cross-Provider Compatibility</strong> Still need format conversion at the client</p><h3>The Key Insight: MCP’s Format IS Anthropic’s Format</h3><p>MCP was created by Anthropic. The tool schema format in MCP is <strong>identical</strong> to Claude’s tool definition format:</p><pre>// MCP Server returns this (which IS Anthropic&#39;s format)<br>{<br>  &quot;name&quot;: &quot;get_weather&quot;,<br>  &quot;input_schema&quot;: {<br>    &quot;type&quot;: &quot;object&quot;,<br>    &quot;properties&quot;: { &quot;city&quot;: { &quot;type&quot;: &quot;string&quot; } }<br>  }<br>}</pre><p>When OpenAI and Google “adopted MCP,” they agreed to <strong>connect to MCP servers</strong> — but their models still expect tools in their own formats.</p><h3>5. The Format Conversion Reality</h3><p>Here’s what actually happens when you use MCP with different providers:</p><h3>OpenAI Expects This Format</h3><pre>{<br>  &quot;type&quot;: &quot;function&quot;,<br>  &quot;function&quot;: {<br>    &quot;name&quot;: &quot;get_weather&quot;,<br>    &quot;parameters&quot;: {<br>      &quot;type&quot;: &quot;object&quot;,<br>      &quot;properties&quot;: { &quot;city&quot;: { &quot;type&quot;: &quot;string&quot; } }<br>    }<br>  }<br>}</pre><h3>Google Gemini Expects This Format</h3><pre>{<br>  &quot;function_declarations&quot;: [{<br>    &quot;name&quot;: &quot;get_weather&quot;,<br>    &quot;parameters&quot;: {<br>      &quot;type&quot;: &quot;object&quot;,<br>      &quot;properties&quot;: { &quot;city&quot;: { &quot;type&quot;: &quot;string&quot; } }<br>    }<br>  }]<br>}</pre><h3>The Conversion Flow</h3><pre>┌─────────────────┐         ┌─────────────────┐         ┌─────────────────┐<br>│   MCP Server    │         │   MCP Client    │         │   LLM Provider  │<br>│   (Confluence)  │         │   (Your App)    │         │   (OpenAI/etc)  │<br>│                 │         │                 │         │                 │<br>│  Returns tools  │────────▶│  CONVERTS to    │────────▶│  Receives in    │<br>│  in MCP format  │         │  provider format│         │  native format  │<br>│  (=Anthropic)   │         │                 │         │                 │<br>└─────────────────┘         └─────────────────┘         └─────────────────┘</pre><p><strong>The conversion burden is on the client side.</strong> If you use Claude, no conversion needed. If you use OpenAI or Gemini, your client (or framework like LangChain) must convert.</p><h3>6. Why Do Different Providers Have Different Formats?</h3><p>This is a historical and technical reality:</p><ol><li><strong>OpenAI introduced function calling in June 2023</strong>–18 months before MCP existed — with their own format (parameters)</li><li><strong>Anthropic created their format</strong> when they launched Claude’s tool use</li><li><strong>Google created their format</strong> for Gemini independently</li><li><strong>The format is tied to model training</strong> — each model was fine-tuned to recognize its specific format structure</li></ol><p>When these companies “adopted MCP,” they adopted the <strong>connectivity protocol</strong>, not a new tool schema format. Changing their formats would require retraining their models — not just updating an API spec.</p><h3>7. Why Don’t MCP Servers Create Provider-Specific Variants?</h3><p>You might wonder: why doesn’t Confluence create separate MCP servers for OpenAI, Google, and Anthropic formats?</p><p><strong>This would defeat the entire purpose of MCP:</strong></p><p>Approach Server implementations needed Without MCP N servers × M providers = explosion With MCP N servers × 1 format = N With variants N servers × M formats = back to explosion</p><p>The trade-off MCP chose:</p><ul><li><strong>Server developers</strong>: Build once in MCP format</li><li><strong>Client/framework developers</strong>: Build conversion layer once per provider</li><li><strong>End developers</strong>: Use a framework that handles conversion (LangChain, LiteLLM, etc.)</li></ul><h3>8. Industry Adoption: What Changed in 2025</h3><p>When I first analyzed MCP, a key question was whether competitors would adopt Anthropic’s protocol. That question has been answered:</p><ul><li><strong>March 2025</strong>: OpenAI announced MCP support in their Agents SDK, ChatGPT desktop app, and Responses API</li><li><strong>April 2025</strong>: Google DeepMind announced MCP support for Gemini models and SDK</li><li><strong>The ecosystem</strong>: Thousands of MCP servers now exist for tools like GitHub, Slack, Postgres, and more</li></ul><p><strong>MCP has become the de facto standard for tool connectivity</strong> — even if the underlying tool schema formats remain provider-specific.</p><h3>9. Conclusion: A Pragmatic Assessment</h3><h3>What MCP Gets Right ✓</h3><ul><li>Recognizes that tool management needs structure</li><li>Provides a clean server/client separation</li><li>Offers useful SDK conveniences</li><li>Creates an ecosystem of shareable, reusable tools</li><li>Achieved industry adoption from major players</li></ul><h3>What MCP Doesn’t Solve ✗</h3><ul><li>Doesn’t eliminate provider-specific tool formats</li><li>Doesn’t create true schema-level standardization</li><li>Conversion still required for non-Claude providers</li></ul><h3>For Developers</h3><p><strong>If you’re building on Claude exclusively</strong>: MCP provides significant value through its ecosystem and native compatibility.</p><p><strong>If you’re building multi-provider applications</strong>: You’ll use MCP for tool connectivity, but still rely on frameworks (LangChain, LiteLLM) for format conversion — or write conversion layers yourself.</p><p><strong>You can absolutely build your own “MCP-like” system</strong> with Spring Boot, Express, or any REST framework. The core concepts — tool discovery and execution — are simple HTTP patterns. MCP adds convention, ecosystem, and industry buy-in, not fundamental new capability.</p><h3>The Real Innovation</h3><p>The breakthrough is <strong>tool calling itself</strong> — the ability for AI models to interact with external systems dynamically. MCP is the <strong>connectivity standard</strong> that emerged to organize that capability. It solved the N×M connectivity problem, but the M×1 format conversion problem still exists at the model layer.</p><h3>Final Thought</h3><p>MCP is useful and has achieved real industry adoption. But understanding what it actually standardizes (connectivity) versus what it doesn’t (tool schema formats) helps you make informed architectural decisions.</p><p>Sometimes a simple REST API is all you need. Sometimes the MCP ecosystem saves you significant work. The right choice depends on your specific use case, provider strategy, and whether the existing MCP server ecosystem covers your needs.</p><p><em>What’s your take? Are you using MCP in production? Have you built alternative approaches? Let’s discuss.</em></p><p><em>Sangeeth K Sivakumar</em></p><p><a href="https://www.linkedin.com/in/sangeethksivakumar/">https://www.linkedin.com/in/sangeethksivakumar/</a><br><a href="https://sangeeth.page/">https://sangeeth.page/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9d35f55c50d1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Zero to Portfolio: Building My First Static Site with ChatGPT and Next.js]]></title>
            <link>https://medium.com/@sangeethksivakumar/from-zero-to-portfolio-building-my-first-static-site-with-chatgpt-and-next-js-b66e9c59d461?source=rss-27480e9e1ebf------2</link>
            <guid isPermaLink="false">https://medium.com/p/b66e9c59d461</guid>
            <category><![CDATA[prototyping]]></category>
            <category><![CDATA[front-end-development]]></category>
            <category><![CDATA[portfolio-website]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[nextjs]]></category>
            <dc:creator><![CDATA[Sangeeth Sivakumar]]></dc:creator>
            <pubDate>Thu, 19 Dec 2024 14:13:47 GMT</pubDate>
            <atom:updated>2024-12-19T14:21:42.669Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Introduction</strong>:</h3><p><em>As a tech enthusiast constantly exploring new tools and frameworks, I had </em><a href="https://sangeeth.page/"><em>my portfolio page</em></a><em>, created with Jekyll and running </em><a href="https://sangeeth.page/"><em>live</em></a><em>. However, I wanted to update it with a fresh, new version built entirely from scratch using Next.js — Tailwind CSS. The idea was to create and design my local version to be clean, functional, and visually appealing, highlighting my skills, projects, and tech journey. With no prior experience in Next.js, I wasn’t sure where to begin. Instead of getting stuck, I decided to collaborate with ChatGPT — my AI coding buddy — and what followed was an incredible journey of learning and building, step by step. Here’s how I went from zero to a fully functional static website in just a few hours!</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O2nPPyvPwQvxOZB6MTlCWA.png" /><figcaption>Existing Live Page</figcaption></figure><h3><strong>How I Started</strong>:</h3><ul><li>The <strong>first prompt</strong> I gave:</li></ul><p><strong>GPT response:</strong></p><p>Initialization steps using create-next-app.</p><p>Include the following:</p><pre>Command: npx create-next-app my-portfolio</pre><ul><li>File structure generated.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5QdU0iRuqef8IHaFZo30Zw.png" /></figure><h3><strong>Building Pages — Some of the example prompts</strong></h3><p><strong>→ About Me Page</strong>:</p><p><strong>Prompt:</strong> “I want a clean home page layout to display my About Me content.”</p><p><strong>GPT Response:</strong> Created page.tsx with &lt;Head&gt; and a clean structure.</p><p><strong><em>What I Tweaked:</em></strong></p><p>“I asked GPT to add emojis and better typography to make it engaging.”</p><p><strong>→ Skills &amp; Accomplishments Section</strong>:</p><p><strong>Prompt: </strong>“Help me display my skills in a grid layout using Tailwind CSS.”</p><p><strong>GPT Response:</strong> Created a responsive grid for skills.</p><p><strong>My Input:</strong> “Add more detailed categories like Full-Stack, Cloud, Mobile, and DevOps.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pP8ObLCxH1MCk1vvlco7WA.png" /><figcaption>Home Page — About me</figcaption></figure><p><strong>→ Sidebar and Social Media Links</strong>:</p><p><strong>Prompt:</strong> “Add a sidebar with navigation links and social media icons.”</p><p><strong>GPT Response:</strong> Generated SVG icons for Twitter, GitHub, LinkedIn, and others.</p><p><strong>My Input:</strong></p><p>• “Align icons into two rows for better spacing.”</p><p>• “Replace Twitter with the new X icon.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/652/1*0k2JOtxL_lUhALToOKFdVg.png" /><figcaption>Side Bar</figcaption></figure><h3><strong>Overcoming Challenges</strong></h3><p><strong>Compilation errors:</strong></p><ul><li><strong>Unescaped Single Quotes</strong> (I’m)</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CZMUe8Ok-j5b3-CRODQt5w.png" /></figure><p><strong>→ Fix shared:</strong> Replace ‘ with &amp;#39;.</p><p>• <strong>Image Optimization Issue</strong>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-48SvNEzlbDnR_DLWY8hrw.png" /></figure><p><strong>→ GPT Fix:</strong> Add images: { unoptimized: true } in next.config.js.</p><h3><strong>Using GPT in VS Code</strong></h3><p>ChatGPT can now read content from your coding apps, bringing you smarter, more accurate answers tailored to your work.</p><p><a href="https://help.openai.com/en/articles/10119604-work-with-apps-on-macos">https://help.openai.com/en/articles/10119604-work-with-apps-on-macos</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c6EzF_i2dToe4hy8G2iWfg.png" /><figcaption>GPT in VS Code assisting with live edits.</figcaption></figure><h3><strong>Final Output and Deployment</strong></h3><p><strong>Prompt:</strong> “How do I convert this Next.js site to static HTML and deploy it to GitHub Pages?”</p><p>• Steps provided:</p><p>• Added output: ‘export’ in next.config.js.</p><pre>Command: npm run build &amp;&amp; npm run export.</pre><p>• Deployed to <strong>Cloudflare Pages</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HmiciCREXlxfaUT4bIvLOA.png" /><figcaption>Final Custom built one (https://sangeeth.page) — Next.js / Tailwind</figcaption></figure><h3><strong>Conclusion</strong></h3><p><em>“I started with no knowledge of Next.js and ended up with a fully functional site. This collaboration taught me the power of GPT in breaking complex tasks into achievable steps. Encourage others to try GPT chat for their projects.”</em></p><p>Happy Coding!! <br>Reach me @<a href="https://www.linkedin.com/in/sangeethksivakumar/">https://www.linkedin.com/in/sangeethksivakumar/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b66e9c59d461" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploring ARM-based cloud compute instances]]></title>
            <link>https://medium.com/@sangeethksivakumar/exploring-arm-based-cloud-compute-instances-e761ef1591fb?source=rss-27480e9e1ebf------2</link>
            <guid isPermaLink="false">https://medium.com/p/e761ef1591fb</guid>
            <category><![CDATA[cloud]]></category>
            <category><![CDATA[arm64]]></category>
            <category><![CDATA[cheap-cloud-hosting]]></category>
            <category><![CDATA[graviton]]></category>
            <category><![CDATA[compute-instance]]></category>
            <dc:creator><![CDATA[Sangeeth Sivakumar]]></dc:creator>
            <pubDate>Sun, 30 Oct 2022 12:41:52 GMT</pubDate>
            <atom:updated>2022-10-30T12:41:52.642Z</atom:updated>
            <content:encoded><![CDATA[<p>#Graviton #AmpereAltra #AppleSilicon #TuaT2a #Arm64</p><h3><strong>My obsession with Arm — Entering Apple M series!</strong></h3><p>I bought the Mac Mini M1 in 2020 which was the transition time for several popular developer tools to support ARM64 architecture in macOS. Incredible to see that it did not take a year also to see most of the packages from Homebrew to the docker engine available for Apple Silicon. By 2022, Apple M series processors became so popular and widely used. Since then, Apple M Series Arm-based processor has become the de-facto one at my home. It has a butter-smooth performance, is noise-free, no heating issues and I can keep going with its pros. Just go for it!! No regrets!!</p><h3><strong>Search for Cloud Compute Instance</strong></h3><p>It started with my personal hobby project work. As a full-stack developer, I just wanted to explore, and learn modern cloud-native tools &amp; patterns. With a basic web service ready locally, I was looking for a cloud-based compute instance. There are several trial services and free credits available from different providers such as Digital Ocean, AWS, Linode, etc. However, my personal hobby project is not just for learning, I wanted to host it and keep it alive for a long time. I didn’t want to spend hundreds of dollars as I am not doing any business, neither there is any RoI. I am a certified AWS Solution architect too and hence my obvious first choice is to go with AWS EC2 which is comparatively expensive to run beyond the trial period.</p><p>While searching for the cheapest(~ $10 pm ), yet reliable cloud compute instances, there were few popular options.</p><p><strong>Choice1:</strong></p><p>EC2 T3/T4 Micro -Very low resource, ok for learning, but really not usable for long-term deployment.</p><p><strong>Choice2 — selected:</strong></p><p>Other popular options like Linode, Upcloud, Digital Ocean, etc. were there. More or less the same fixed price, and it was ok to start with. So, I chose the most popular one — Digital Ocean<strong> Droplets.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K0B0wtxJVzZPo_BODu0LEg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yiq9G8sWFyqcWXNEJxw5mg.jpeg" /></figure><p>Signed up! Added credit card, Happy purchase!!.</p><h3><strong>The elephant in the room: X86 vs Arm64</strong></h3><p>I had my container images generated locally and stored in GitLab&#39;s &amp; docker hub. I used both kubectl (K8) &amp; docker compose to deploy! Boom — Failed!! There revealed the obvious elephant in the room. I was happily creating docker images from my local Apple M1 system, which is Arm64 based. The Digital ocean droplet I purchased was Intel X86 based. Then, I went through the docker documentation and found the ‘docker buildx’, a tool to build multi-architecture docker images. I have made a separate blog on how to use it <a href="https://www.sangeeth.page/posts/docker-image-multi-architecture-arm64/">here</a>. But, the availability of matured &amp; verified base images built and published specifically for Arm64 is very less compared to X86 based.</p><p>This workaround was ok, but there were several instances where the buildx command failed, and I always had to google and fix it.</p><h3><strong>Back to AWS:</strong></h3><p>Digital ocean was ok, but as an AWS developer, I wanted to use different seamless integration options within AWS such as SES, Cognito, gateway, CF etc. Realized that the low-cost digital ocean is ok only for basic web projects. I wanted some solution for the long term and hence checked for a few savings general purpose instance plans in AWS.</p><p>t4g.small — 2GB, 2Vcpu, $0.0134/hr , 10$ / pm<br>t3.small — 2GB, 2 Core, $0.0166 / hr, 12.55$/ pm</p><p>Wow!! Graviton( t4G) made my life easy. The advantage is not about the negligible difference in cost per month. This graviton instance is arm64 based. Yaayyy!!!! Now, I can simply build from my Apple Mac M1 and deploy it straight into AWS EC2. No more workaround.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/586/1*o9dl8Vv8JrMS3mIszHm28w.jpeg" /></figure><h3><strong>Benefits of Graviton Processor</strong></h3><p>Is cost the only deciding factor? No!! There are more reasons. One of the highlights of any Arm-based processor is its energy efficiency (same as why my Mac M1 is super silent and so our powerful smartphones).</p><p>As per AWS documentation,</p><ul><li>Up to 40% better price performance over comparable current generation x86-based instances.</li><li>Graviton3-based instances use up to 60% less energy for the same performance than comparable EC2 instances.</li><li>Reduce their carbon footprint</li><li>Cost Efficient</li></ul><h3><strong>How about a Performance Benchmark?</strong></h3><p>I found some interesting article that shows better performance while using arm64 instances. Eg. below is from, <a href="https://www.tigera.io/blog/is-arm-architecture-the-future-of-cloud-computing/"><strong>Tigeria</strong></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/616/1*6cqsZ7po1WYNLaKISNaf3w.jpeg" /><figcaption>Image from Tigeria</figcaption></figure><p><a href="https://www.instana.com/blog/the-future-is-now-arms-aarch64-on-the-rise-with-instana/"><strong>Instana</strong></a>, an <strong>IBM</strong> company also talks about their support for arm64 and how the future will be arm64.</p><h3><strong>Mac Mac Mac. How about Windows? — Project Volterra</strong></h3><p>So, the usage and support of ARM-based processors are not just limited to smartphones and the Apple ecosystem. Even Microsoft knows the future and hence unveiled the Windows Dev Kit 2023 a.k.a. <a href="https://blogs.windows.com/windowsdeveloper/2022/10/24/available-today-windows-dev-kit-2023-aka-project-volterra/"><strong>Project Volterra</strong></a> , an Arm-powered device built by Windows developers for Windows developers.Ref: <a href="https://www.microsoft.com/en-us/d/windows-dev-kit-2023/94K0P67W7581?activetab=pivot:overviewtab"><strong>https://www.microsoft.com/en-us/d/windows-dev-kit-2023/94K0P67W7581?activetab=pivot:overviewtab</strong></a></p><h3><strong>Other popular ARM instances</strong></h3><ul><li><strong>Oracle</strong> used Arm Ampere A1 Compute instance, which offers generous <a href="https://www.oracle.com/sg/cloud/free/"><strong>free</strong></a> hour usage.</li><li><strong>Google</strong> Cloud launched its ARM-based compute recently — <a href="https://cloud.google.com/compute/docs/instances/arm-on-compute"><strong>Tau T2A</strong></a></li><li><strong>Azure</strong> launched <a href="https://azure.microsoft.com/en-us/blog/azure-virtual-machines-with-ampere-altra-arm-based-processors-generally-available/"><strong>Ampere Altra</strong></a> Arm-based compute instance earlier this year and is now available generally.</li></ul><p>It is very exciting to see the transition in the enterprise and clearly, with hybrid, edge computing, IoT, and 5g demands, the energy need is increasing rapidly. ARM-based cloud instances would definitely take over the traditional AMD &amp; Intel X86 processors.</p><p>Let me know your thoughts:)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e761ef1591fb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>