First Came The Tokenizer
A deep dive into tokenizers, the invisible first piece of your LLM stack. Learn how they control costs, context windows, and performance, and see how algorithms like BPE and SentencePiece can make or break your AI.
Osman's Odyssey: Byte & Build
Recent content on Osman's Odyssey: Byte & Build
Latest posts
So You Want to Learn LLMs? Here's the Roadmap
The straight-up, no-BS roadmap for learning LLMs in 2025. Skip the ML fluff and endless prerequisites. Get the actionable phases, projects, and resources to actually build, train, and ship large language models—from the ground up.
Software Engineers Aren't Getting Automated—Local AI Has To Win
Stop worrying about AI replacing you—the real threat is losing technical depth. As cloud dependence grows and platforms get more opaque, local-first AI, open weights, and full-stack ownership are the only safety nets left. Why the future belongs to those who can build, debug, and own their tools from the metal up. Trust no corporate overlord.
My Ultimate DeepResearch Prompt Builder Template and How I Use It
I’m sharing my DeepResearch prompt builder template—the system that powers my research and learning workflows. Learn exactly how I turn chaos into clarity, force actionable insights, and get the most out of LLMs. See the template, my step-by-step process, and real-world tips for DeepResearchMaxxing in 2025.
Just Like GPUs, We Need To Be Stress Tested
Why 101 days of daily tech blogging? A raw, open challenge on AI, LLMs, self-hosted experiments, knowledge distillation, and why consistency beats talent. Expect rants, technical breakdowns, open hardware journeys, memes, and daily accountability from the basement AI server guy.
Mastering the Game: How Corporate Politics Shape Your Career
Corporate politics isn't just backroom deals—it's how influence, visibility, and relationships shape your career. In this candid guide, you'll learn to use titles, politics, and intentional networking to your advantage (without selling your soul). Real talk from people who've played—and won—the game at big tech and beyond.
Once Undesirable, Now Undeniable
How taking risks, building in public, and refusing to play a losing game flipped the script—and why sometimes you have to become undeniable before you ever become accepted.
Build Your Private AI Screenshot Organizer with LMStudio
Build a local, privacy-first screenshot organizer using LMStudio’s Python SDK and Gemma 3 multimodal models. Keep your data off the cloud, automate screenshot categorization, and leverage the power of open-source AI—all running from your own PC. Step-by-step guide, code walkthrough, and a practical use-case for local LLMs.
No, RAG Is NOT Dead!
Forget the hype—here’s what actually happened when we asked “Is RAG dead?” This deep-dive explores why Retrieval-Augmented Generation (RAG) is still essential in real AI systems, what people get wrong, and how practitioners are shipping the next wave of AI with smarter retrieval, dynamic context, and hard-earned lessons from the field.
From the Shadows to the Feed: Why I’m Finally Playing the Game
After years of building in the dark, I decided to play the game of distribution. Here’s why networks—and distribution—matter more than ever, and why I’m finally sharing my journey, experiments, and ideas in public.
Key Highlights From Running DeepSeek R-1 671B on 14x RTX 3090s + Epyc 7713 & 512GB RAM
Key takeaways from livestreaming DeepSeek R-1 671B (4-bit) on a 14x RTX 3090 basement AI server. See how KTransformers crushed llama.cpp in prompt eval speeds, compare setups, and get real-world insights into massive LLM inference with vLLM, ExLlamaV2, and more.
Stop Wasting Your Multi-GPU Setup With llama.cpp
Exploring the intricacies of Inference Engines and why llama.cpp should be avoided when running Multi-GPU setups. Learn about Tensor Parallelism, the role of vLLM in batch inference, and why ExLlamaV2 has been a game-changer for GPU-optimized AI serving since it introduced Tensor Parallelism.
Resources From X/Twitter Audio Space on LLMs & AI - 2025-02-02
A curated collection of links, books, tools, and benchmarks discussed during the February 2nd, 2025 Twitter/X Audio Space on LLMs and AI. Includes practical resources, RAG leaderboards, toolkits, and perspectives on AI adoption in the Middle East and globally.
Antifragile AI
Explore how AI systems can become antifragile, harnessing uncertainty to thrive. Learn about the shift and acceleration from traditional software to AI agentic systems and their implications for the future.
All In
Embrace new ideas, trust your instincts, and go all in. 42 days to launch—let’s win this game! #GoAllIn
Serving AI From The Basement — Part II
SWE Agentic Framework, MoEs, Quantizations & Mixed Precision, Batch Inference, LLM Architectures, vLLM, DeepSeek v2.5, Embedding Models, and Speculative Decoding: An LLM Brain Dump... I have been working on a multi-agent system that simulates a team of Software Engineers; this system assigns projects, creates teams and adds members to them based on areas of expertise and need, and asks team…
@TheAhmadOsman
Hi There! 👋 Welcome to my corner of the internet where every line of code, every 3D printed layer, and every gym rep is a brushstroke on the canvas of creation. By trade, I’m a software engineer, but really, I’m a perpetual learner; a builder; a problem solver; a creator; a thinker; and a tinkerer in the grand workshop of life. Bit By Bit: The Journey & The Camels Born and raised on the the…
Serving AI From The Basement — Part I
Dedicated LLM server powered by 8x RTX 3090 Graphic Cards, boasting a total of 192GB of VRAM.