Plasma Code: A Machine Learning Hackathon at KPI

A useful warning must arrive early enough to act on. At Plasma Code, a machine-learning hackathon held at Igor Sikorsky Kyiv Polytechnic Institute on October 9–11, 2026, participants explored that problem through a plasma simulation: how to recognize a developing instability without raising too many false alarms.

The Faculty of Physics and Mathematics and Abitly organized the event, bringing together more than 50 participants in ten teams. DataRoot Labs contributed organizational and mentoring support.

Participants came from computing, applied mathematics, physics, radio engineering and materials science backgrounds. The universities represented included KPI and Taras Shevchenko National University of Kyiv.

The event brief called for basic Python and machine-learning knowledge. A specialist background in plasma physics was not required: the program included supporting material, descriptions of the data and access to mentors.

The task was to work with data from a software simulator of plasma behavior. Participants examined numerical signals, tested physical hypotheses and built models that could anticipate a loss of plasma stability. Their reported results concern this simulated environment.

The organizers' recap describes three approaches among the prize-winning work:

These questions lead to different experiments. A model that detects an event just before it happens may leave little time to respond. An earlier warning is harder because the signals are less conclusive. Raising the alarm more readily can catch more events while also raising false alarms on simulated runs that would have remained stable.

That made the work about more than a single accuracy figure. Participants had to examine the assumptions behind a model, decide when it should raise an alarm and explain the trade-offs in their results. The program included consultations with mentors and presentations to an expert jury.

These are educational project results within the hackathon's software environment, rather than validation of a system on physical equipment.

The format connects technical learning with a concrete problem: participants can inspect signals, form a hypothesis, test it and explain why a result changes.

A selection of photographs from Abitly's event album. Use the arrows or thumbnails to browse.

The KPI event report provides the university's account, while the original announcement sets out the challenge and preparation expected of participants. Our AI in Chemistry report explores another setting where scientific knowledge helps shape model inputs, experiments and evaluation.