What it does This is a sprawling, open-source textbook-in-markdown covering the full stack an AI/ML research engineer actually needs: linear algebra and calculus, classical ML, NLP, computer vision, speech, multimodal models, GNNs, SIMD/GPU programming, inference optimization, and production MLOps. The author, Henry Ndubuaku, compiled it from years of personal notes; friends reportedly used it to prep for interviews at DeepMind, OpenAI, and Nvidia. It ships with an MCP server so Claude, Cursor, or VS Code can query the material directly as a knowledge base.

The interesting bit The scope is deliberately anti-silo. Most resources stop at “learn PyTorch” or “crack the coding interview.” This stitches together the math, the model architecture, the low-level kernel, and the serving infrastructure as one continuous narrative. The foreword is also uncommonly candid about learning as a skill—complete with a Charles Darwin analogy and a Serbian IQ study.

Key highlights

• 18 chapters available now, from vector spaces to ML systems design; 2 more (Applied AI, Bleeding Edge AI) marked “Coming”
• Includes modern topics often missing from textbooks: flow matching, SSMs, speculative decoding, ARM NEON/SME2, WebGPU
• MCP server integration for AI-assisted study workflows
• Explicitly targets “intuition first” explanations over dense notation
• Requires only elementary math and basic Python as prerequisites

Caveats

• The “friends got into DeepMind/OpenAI/Nvidia” claims are anecdotal, not independently verifiable
• Two final chapters are still marked “Coming”; the repo is a living draft
• The MCP server requires a local clone, so it’s not a plug-and-play web service

Verdict Worth bookmarking if you’re self-teaching ML and tired of jumping between a math textbook, a PyTorch tutorial, and a systems blog. Skip it if you want a polished, peer-reviewed reference or a quick interview cheat sheet—this is a marathon, not a sprint.