A Reinforcement Learning tutorial with no math: just code for programmers

On 6 July 2026, Habr published a tutorial on reinforcement learning (Reinforcement Learning). Its author, from Cinimex, set out to fill a long-standing gap: explaining RL to programmers through clean code, with mathematical derivations left out entirely.

Why most ML tutorials are hard to read

Most ML learning material is written by people with a math education, which is no surprise. Data science and ML engineers typically come into the profession through math and statistics, not through programming. Their code reflects that mindset: it is structured like a mathematical proof and packed with terms and abstractions that only make sense to those who already know the theory.

For an experienced programmer without deep mathematical training, such code is unreadable, and the problem is not syntax. Each line makes sense on its own, but why they run in this order with these parameters is unclear. The complexity of the language constructs is layered on top of a demanding theoretical base. As a result, even a conscientious tutorial turns into a set of formulas whose meaning cannot be reconstructed without prior knowledge.

"If you are not familiar with the theory, it can sometimes be simply

impossible to guess from the code why the actions are being performed," the author explains.

The paradox is that programmers are the largest audience interested in applying ML in real products. Yet educational content written from a developer's perspective rather than a researcher's is still extremely scarce.

What has changed since 2019, and what hasn't

The author became interested in ML and AI in 2019, when the topic stopped being the preserve of academic science and entered the technology mainstream. Since then, the number of articles and code examples available openly has grown many times over: courses in Russian, topical channels and communities with thousands of members have appeared.

But one thing has stayed the same: the coding style of the examples and how math-heavy they are. There is more content, but the barrier to entry for a programmer without serious mathematical training is unchanged.

The tutorial takes a different approach:

• no math formulas, only code
• each action is explained in terms of what it means, not derived from a theorem
• the writing style is closer to real projects than to academic papers
• the material is aimed at readers with school and university-level math behind them

Explaining through production code became the standard in web development long ago, but in ML it is still the exception rather than the rule.

Why Reinforcement Learning in particular

Reinforcement learning is one of the hardest areas of ML. An agent learns not from labeled data but by interacting with an environment: it tries actions, receives a reward or penalty signal, and learns to maximize long-term reward. RL underlies RLHF (Reinforcement Learning from Human Feedback), the key method used to align modern large language models, including GPT and Claude.

That makes understanding the basic principles of RL useful beyond researchers. Developers who work with AI systems and integrate language models into products gain a deeper understanding of what they work with every day.

Yet good "for programmers" explanations of RL are still almost nonexistent, in either Russian or English. Most introductory material is either too academic or too superficial. The Habr tutorial aims for something in between: depth without the math barrier.

What this means

Educational ML content is gradually moving toward an engineering audience. Authors who can explain mathematically complex concepts through code without losing the meaning are few. If the tutorial holds to its stated style throughout, it could become one of the most useful resources for Russian-speaking developers taking their first steps in ML.