Start with intuition, before any code
Before touching a line of code, it's worth spending a few hours on visual, intuitive explanations of what a neural network is actually doing — 3Blue1Brown's neural network video series is the most commonly recommended starting point for exactly this reason, and it's free on YouTube.
Google's own Machine Learning Crash Course (developers.google.com/machine-learning/crash-course) is a free, roughly 15-hour self-study course built for exactly this stage: no advanced math required, just basic Python and algebra, with interactive visualizations rather than a wall of theory.
Get hands-on, then go deeper
Kaggle Learn's short, free micro-courses get you writing actual code and submitting real predictions within hours, not weeks — a faster path to "I built something" than a long theoretical course.
fast.ai's Practical Deep Learning for Coders (course.fast.ai) takes the opposite approach from most university-style courses: you train a working image classifier in lesson one, then work backward into the theory as you need it — free, and aimed at people who already have roughly a year of coding experience.
Specialize and join a community
Once you have the basics, Hugging Face's free courses (huggingface.co/learn) go deep on the specific area most people actually want now — the NLP/LLM course, plus dedicated free courses on AI agents and diffusion models, all using the same open-source tools the field actually runs on.
For Russian-speaking learners, the Yandex School of Data Analysis (ШАД) publishes its full machine learning handbook and course materials for free at education.yandex.ru/handbook/ml — genuinely rigorous, university-level material, not a marketing funnel. Joining a community (ods.ai is the largest Russian-language ML community) matters more than it sounds: a real project with feedback from people who've done it before catches mistakes no course alone will.