Presentations

Systemic training of Data Science specialists: educational architecture for the transition from low-code to full-code

Kondrashova E.S., Milovidova A.A.

Federal State Educational Institution of Higher Education “Dubna State University”, kes.th@uni-dubna.ru

Demand for Data Science specialists is growing at 25–30% annually. The traditional model based on Python creates high entry barriers and cognitive overload, while low-code platforms deliver rapid results but are limited in flexibility. A contradiction arises: how can one combine both approaches? Most programs choose either theory or practice, failing to integrate them into a unified architecture.

The methodology rests on Cognitive Load Theory, demonstrating that low-code environments reduce syntax burden, and the Zone of Proximal Development principle, confirming that tools expand student capabilities. The spiral model implies revisiting material with increasing complexity.

An architecture with three competency levels is proposed: visual, hybrid, and programmatic. The key principle is analytical logic invariance—the same task has equivalent representation in low-code and full-code formats. This strengthens understanding and reduces the time to establish foundational competencies by 30–40%.

The architecture comprises five components. The objectives component defines competencies: analytical thinking, data preparation, exploratory analysis, modeling, and interpretation. The content component organizes material using a spiral model. The technology component specifies the sequence: Orange, then KNIME or Loginom, then Python. The organizational component structures lessons according to the analytical cycle: problem formulation, exploration, data preparation, modeling, evaluation, and interpretation. The assessment component implements three levels: formative assessment, checkpoints, and project portfolios.

The transition occurs through five phases: Phase 1—100% low-code, Phase 2—75% low-code and 25% code, Phase 3—parity of approaches, Phase 4—code predominance, Phase 5—full transition to full-code development in Python.

Practical implementation is based on dual representation: each task is first solved in a low-code environment, then reproduced in Python, demonstrating analytical logic invariance and facilitating transition to programmatic thinking.

The architecture resolves the central contradiction in Data Science education: combining rapid achievement of practical results with deep programming skills. The approach is applicable to university programs and corporate training.

• abstract in Russian (PDF)