ChatGPT and Gemini in tester training: a Stepik course author's experience with beginners

Artem Rusov, author of testing courses on Stepik and Udemy, explained how he builds AI into training for beginners. In his experience, models do not replace the classic course but extend it: they help create custom training environments, generate tasks and give students livelier practice.

A custom training sandbox

The turning point for the author came in 2023, when, without a deep programming background, he used ChatGPT to build a full sandbox for his main course. This is not a set of mocks but a system with a frontend, backend, documentation, mockups and a database. For a manual-testing teacher, this turned out to be more than a flashy demonstration of what AI can do: it gave him an environment that fully matches the course's logic and does not depend on someone else's limitations.

The approach quickly showed practical benefits. The teacher can plant the bugs he wants, fix discovered problems quickly and build scenarios that actually test a skill, rather than the ability to get through a specific practice tool. According to Rusov, the student then works not in a sterile learning shell but in a system that resembles a real product. The course author gets an extra bonus too: growing expertise. Building a learning service yourself means understanding code, infrastructure and delivery processes much better.

Task generation and RAG

The next stage is moving AI from course preparation directly into the user experience. Rusov is now developing a new version of practice tools in Gemini, where a student will be able to generate any number of new tasks in the same interface and from the same template. This is no longer a one-off experiment: to keep everything working under a large number of requests, you have to think about API keys, cloud platforms and system stability under load.

• A custom sandbox built for the course program
• Browser-based practice tools for requirements, SQL, DevTools and bug hunting
• Tasks graded by an LLM with automatic feedback
• RAG search over a verified knowledge base instead of random generation
• Generated illustrations and a story wrapper for the course

In parallel, the author is learning RAG so that students look for answers not in the model's arbitrary output but in a pre-selected knowledge base. By the same logic, the courses now include tasks where AI evaluates a detailed answer, assigns a score and leaves a comment. This tool, developed together with a colleague, plugs into the educational platform via the admin panel and an iframe. But Rusov sets out the limits of the approach right away.

"In education, the key is to do no harm."

Gamification without blind faith

Another direction is the visual and narrative packaging of the course. The author writes that he turned his longtime mascot, an artichoke, into a character named Arty, who travels through ancient civilizations and tests APIs. For this setting he builds separate sandboxes and images generated with Nana Banana. The idea is not decoration for its own sake, but to make the course more memorable and make it look like a cohesive product rather than a set of disconnected lessons and tests.

Rusov does not set AI against classic teaching formats. Ordinary tests, matching tasks, chronology reconstruction and fill-in-the-blank exercises are not going away, and he reads through, edits and fact-checks by hand all texts created with the model. His main thesis is simple: you cannot mass-generate "AI slop", because the author is responsible for every piece of the course. In this setup, AI is not a replacement for methodology but a tool that works only within clear boundaries and editorial control.

What this means

Rusov's case shows that AI in online education starts to deliver value not at the level of slogans but in concrete product elements: from sandboxes and answer checking to personalized practice tools and the visual layer. But the gain appears only where the course author retains responsibility for content, quality and teaching logic, rather than shifting everything onto generation.