DR.VOLODYMYRSOKOL

CHALLENGE

Traditional university curricula often leave a gap between academic knowledge and the skills companies actually hire for. Students graduate with theory but limited project experience, and companies struggle to find job-ready juniors.

SOLUTION

Founded in 2018 and integrated into Bachelor-level computer science education, Innovation Campus replaces rigid tutorials with challenge-based and project-based learning. Students follow individual learning paths, work on industry-contributed projects, and are supported by peer assessment, gamification, automated checks, plagiarism detection, and learning analytics. The ecosystem continued operating online during COVID-19 and the war in Ukraine.

OUTCOME

Serves more than 200 learners per year and received the Silver Award for Europe at the Wharton-QS Reimagine Education Awards 2022.

CHALLENGE

Recognizing international academic records requires comparing course content across languages, structures, and grading systems — a slow, expert-driven process that is hard to scale.

SOLUTION

A supervised pipeline extracts course data from documents with PyMuPDF, computes semantic similarity with SBERT and Azure OpenAI components, and combines engineered features with logistic regression. Results surface in a web interface designed for human-in-the-loop decision support, keeping the final judgment with the expert.

OUTCOME

In the supervised thesis evaluation on 52 historical cases, the best configuration reached 85.9% accuracy.

Result from a supervised thesis evaluation on 52 historical cases — not a production-wide guarantee.

CHALLENGE

With hundreds of possible learning challenges, students need help finding the shortest valid path to a target competency — one that respects prerequisites and matches their demonstrated performance.

SOLUTION

Courses and challenges are modeled as a graph. Mixed-integer programming finds efficient paths, K-nearest-neighbor estimation predicts grades from student-performance vectors, and pgvector on PostgreSQL powers similarity search behind a REST API. The approach was validated with synthetic-data experiments.

OUTCOME

Published as a peer-reviewed architecture (Journal of Physics: Conference Series, 2025) and forms the recommendation backbone for challenge-based curricula.

CHALLENGE

Medical visualization environments need to know which surgical instruments are present in the frame — reliably and in real time — without manual tagging.

SOLUTION

A YOLOv5-based detection model, built with PyTorch alongside TensorFlow and Keras tooling, detects and classifies surgical equipment with real-time inference and bounding-box visualization inside the target environment.

OUTCOME

A working real-time detection prototype developed under research and thesis supervision, demonstrating applied computer vision in a medical visualization context.

CHALLENGE

Scaling honest, high-quality assessment is one of the hardest problems in technical education: open answers need judgment, code needs execution and originality checks, and instructors need insight into how learning actually progresses.

SOLUTION

A set of components built around Azure OpenAI with guardrails: open-answer assessment support, automated exercise correction for programming tasks, source-code similarity and plagiarism detection, and learning-analytics reporting — integrated through REST APIs with Jupyter Notebooks and Open edX.

OUTCOME

Working AI-assisted assessment and analytics components used in research and teaching practice, keeping instructors in control of final grading decisions.