1,417 saved links. Honest corpus note: the tag is a magnet for #MachineLearning-hashtagged saves, so much of the stream is really agentic-AI news (see AI Agents for that). This page curates what’s genuinely ML: the mathematics, the models, the evaluation discipline, and the hands-on fundamentals — the layer beneath everything you’re building.
Related: LLMs · Deep Learning · Data Science · MLOps
Start here
- The three pillars of ML: linear algebra, calculus, probability — with a full roadmap.
- mlcourse.ai — self-paced ML course with Kaggle assignments — the classic structured path.
- Patterns, Predictions, and Actions — the free graduate-level story of ML.
- the free-resources shortlist: HF Course, FastAI, OpenAI Cookbook and more.
- an AI/ML interview study booklet — foundations connected to the modern stack; doubles as a self-assessment.
The mathematics
- Pen and Paper Exercises in Machine Learning — a free 211-page PDF of worked problems (paper version). Nothing builds intuition faster.
- “Linear Algebra Is Not Hard” — a visual, animated open-source book — geometry-first treatment of vectors, matrices, eigendecomposition.
- NPTEL: Applied Linear Algebra in AI & ML.
- Cambridge quietly released 10 free university-level AI/ML textbooks.
- Justin Math: the textbook behind the most advanced high-school math/CS sequence in the US (direct PDF) — culminates in ML from scratch.
- causalai-book.net — causality, the pillar most courses skip.
Under the hood
- The journey of a token: what really happens inside a transformer.
- Raschka: The Big LLM Architecture Comparison — the field’s best model-anatomy survey.
- LLM Embeddings Explained: a visual and intuitive guide and vector databases in 3 levels of difficulty.
- Small models (the counter-trend worth watching): the complete 2026 guide to small language models · FunctionGemma — a 270M function-calling model that runs on your phone · the “Welcome to Gemma 4” launch post · building agents on local small models.
- fine-tuning a model is less complicated than it looks — the HF tutorial.
Evaluation — the discipline that separates demos from systems
The densest quality cluster in the corpus, and the one most relevant to your platform work.
- Cameron Wolfe: understanding LLM evaluation — whether you build LLMs or build with them, this is the skill.
- The LLM Evaluation Guidebook v2 — interactive, comprehensive.
- “If you can’t test it, don’t deploy it” — the new rule of AI development.
- Building evals for AI adoption: from principles to practice.
- evaluating AI agents: real-world lessons from Amazon and measuring whether an agent is hallucinating.
- beyond accuracy: 5 metrics that matter for agents.
- Benchmarking platforms: Kaggle’s LLM eval product · Kaggle Game Arena — models compete head-to-head in strategic games · Google’s LLM-Evalkit for prompt-engineering metrics.
Learning by doing
- deep-ml.com — LeetCode-style ML problems, solved in the browser.
- interactive-ml.com and ml-visualiser.vercel.app — visual, interactive concept explorers.
- the 8-week ML & MLOps accelerator — “most courses stop at the notebook; this one doesn’t” and ml.school.
- Google’s 5-day self-paced agents intensive · Google’s generative-AI fundamentals path.
- Collections: the become-an-AI-engineer-for-free resource stack · tools and datasets for ML · AI/ML roadmap including agentic AI · AI-ML-Cheatsheets · Machine Learning with Python Tutorial · ML with Python and turtle graphics — a visual guide.
Where ML meets the day job
- DORA 2025: the state of AI-assisted software development — ML’s measured impact on your other discipline.
- HBR: AI doesn’t reduce work — it intensifies it — the sober counterweight.
- “Your AI feature isn’t impressive. Your fallback strategy is.” — resilience engineering for model-backed systems.
- regression language models: predicting industrial system performance from raw text — ML applied directly to ops telemetry.
- “ML Kit is designed to detect faces, not recognise people” — a well-reported cautionary tale about deploying models beyond their design intent.