https://ragibcs.gitlab.io/posts/ Recent content in Posts on Ragib CS Hugo en-us Fri, 03 Apr 2026 10:00:00 +0600 https://ragibcs.gitlab.io/posts/building-linfer-rust-llm-inference/ Fri, 03 Apr 2026 10:00:00 +0600 https://ragibcs.gitlab.io/posts/building-linfer-rust-llm-inference/ <h2 id="why-build-another-inference-engine">Why Build Another Inference Engine?</h2> <p>When I started working with large language models locally, I quickly ran into the usual suspects: slow inference times, memory bloat, and dependency hell. Most existing solutions are either too heavyweight (PyTorch with CUDA) or too opinionated about model formats.</p> <p>So I built <strong>linfer</strong> — a Rust-based local LLM inference engine that&rsquo;s 3x faster than comparable solutions for CPU inference.</p> <h2 id="the-core-problem">The Core Problem</h2> <p>Running LLMs locally on CPU is painful:</p> https://ragibcs.gitlab.io/posts/lessons-from-200-competitive-programming-problems/ Wed, 01 Apr 2026 14:30:00 +0600 https://ragibcs.gitlab.io/posts/lessons-from-200-competitive-programming-problems/ <h2 id="the-journey">The Journey</h2> <p>I started competitive programming in 2021 with zero algorithmic knowledge. Fast forward to today: 200+ problems solved across Codeforces, LeetCode, CodeChef, and AtCoder.</p> <p>This isn&rsquo;t a &ldquo;how I became red on Codeforces&rdquo; story. I&rsquo;m still grinding through Div 2 problems. But I&rsquo;ve learned a lot about problem-solving, pattern recognition, and how to think algorithmically.</p> <p>Here&rsquo;s what actually stuck.</p> <h2 id="1-most-problems-are-variations-of-20-patterns">1. Most Problems Are Variations of ~20 Patterns</h2> <p>After solving 50 problems, I thought every problem was unique. After 200, I realized they&rsquo;re all remixes.</p> https://ragibcs.gitlab.io/posts/running-llms-locally-arch-linux/ Sun, 15 Mar 2026 10:00:00 +0600 https://ragibcs.gitlab.io/posts/running-llms-locally-arch-linux/ <h2 id="why-i-built-this-instead-of-using-ollama">Why I Built This Instead of Using Ollama</h2> <p>I wanted to run AI models locally on my Arch Linux laptop — privately, offline, with zero cloud dependency. Ollama seemed like the obvious choice, but I didn&rsquo;t want an opaque, heavy framework. I wanted control.</p> <p>After navigating outdated documentation and breaking changes, I got llama.cpp running with Vulkan GPU acceleration on my Intel Iris Xe. This is the guide I wish I had.</p> https://ragibcs.gitlab.io/posts/cliproxyapi/ Fri, 04 Apr 2025 00:00:00 +0600 https://ragibcs.gitlab.io/posts/cliproxyapi/ <blockquote> <p><strong>Guide based on</strong> CLIProxyAPIPlus <code>v6.9.1-0-plus</code> · Commands may vary in future versions.</p> </blockquote> <p>Connect two powerful AI providers — <strong>GitHub Copilot</strong> and <strong>Kiro AI</strong> — to your self-hosted <a href="https://github.com/eceasy/cli-proxy-api-plus">CLIProxyAPIPlus</a> instance in minutes using OAuth device flow authentication.</p> <hr> <h2 id="prerequisites">Prerequisites</h2> <p>Before you begin, make sure you have the following in place:</p> <table> <thead> <tr> <th>Requirement</th> <th>Details</th> </tr> </thead> <tbody> <tr> <td><strong>CLIProxyAPIPlus</strong></td> <td>Running via <code>eceasy/cli-proxy-api-plus:latest</code></td> </tr> <tr> <td><strong>GitHub Copilot subscription</strong></td> <td>Individual, Business, or Enterprise</td> </tr> <tr> <td><strong>AWS Builder ID or Kiro IDE</strong></td> <td>For Kiro AI login</td> </tr> <tr> <td><strong>Docker</strong></td> <td>Installed and accessible in your shell</td> </tr> </tbody> </table> <p>Verify your container is running before proceeding:</p>