School Students Master Communication with GigaChat at HSE and Sber Hackathon

In late December 2025, a unique competition was held at HSE University where participants solved challenges not by writing code, but solely by interacting with Sber’s GigaChat artificial intelligence model. The Improm(p)tu hackathon was an experiment less about programming skills than a new form of literacy: the ability to work effectively with AI by translating complex problems into a language neural networks can understand.

While submitting a prompt and receiving a response may seem like the simplest thing in the world, Deputy Vice Rector Danil Fedorovykh emphasised that the key to communicating with neural networks is precise prompt engineering. Opening the hackathon, he addressed the students and stressed that the capabilities of AI extend far beyond completing academic tasks, and that honing one’s prompting skills is a way to stand out among peers and advance on the path to professional growth.

This idea was the event’s core theme. While AI is typically banned in academic competitions, here it was an essential tool. For the first time in Russia, hackathon participants had to create a mini-book or script (at their discretion) using only GigaChat—from the initial concept to the final text.

© HSE University

Before the start, Viktoria Permikina, an expert from Sber’s Directorate of Academic Partnerships, addressed the participants: ‘You are not here today to compete with artificial intelligence, nor simply to use it. You are here to learn how to cooperate with it. At Sber, the creator of GigaChat, we see such technologies not just as a smart search or text generator, but as a new environment—digital material your generation will work with. The ability to communicate competently with AI is no longer an optional skill but a fundamental literacy, like reading and writing. Today, you will act as directors for a neural network. By refining your prompting skills, you lay the groundwork for future success in any field, from science and art to business. Don’t be afraid to experiment, learn to formulate ideas clearly, and remember: the most powerful intelligence here is your own. Good luck! May today’s dialogue with GigaChat be your first step towards a great collaboration with AI!’

The competition blurred the lines between a technical hackathon and a humanities-based challenge. Participants didn’t need to know Python but were required to think structurally, consistently, and creatively. Their goal was not merely to receive text from the GigaChat neural network, but to learn how to guide it: to construct a dialogue and adjust the style, length, and content to produce a coherent, substantial, and meaningful piece of work.

The event began with an educational session where a Sber expert introduced participants to prompt engineering fundamentals, from simple queries to complex multi-step techniques. This was followed by nearly four hours of intense practical work, during which each participant engaged in a unique dialogue with AI, iteratively refining prompts to achieve their desired result.

For many, this was their first serious experience using AI not as an academic aid but as a professional tool. They discovered that generating ‘just text’ was easy, but crafting text with a specific purpose, idea, and structure was a challenge requiring significant human intelligence.

© HSE University

The top 10 finalists were those who approached GigaChat not as an omniscient programme but as a powerful, obedient tool. They did not just ask questions; they built strategies, provided context, set clear boundaries, and clarified details. Their prompts resembled collaborative scenarios more than simple queries.

‘I decided to participate in the hackathon because I understand the future is uncertain; a lot will change, but AI will definitely become one of the main tools,’ says Alexander Kirilin, winner in the Script category. ‘I wrote a theatre script, not a film script, because that genre is more familiar to me. I set the story at HSE University: the characters—my classmates and the track mentor—are working at a hackathon. They go through difficulties, break down, pull themselves together, and ultimately win. Knowing GigaChat would evaluate my work, I first asked what topics and styles it considered suitable for each of the examiners. I sent my draft multiple times, asking to critique the text as an examiner would—highlighting weak points and problems. It also helped that I described the TDA and asked it to generate a prompt based on it, which I then sent to another chat. My previous experience in creating long texts with AI also helped me: at our school, writing a social studies research paper is a mandatory requirement for completing the course.’

An important outcome was the participants’ realisation that prompt engineering is less about IT and more about communication, clarity of thought, and the ability to be understood—even by artificial intelligence.