Ірина

Iryna Popovych
Junior ML Engineer & AI Developer
Kosice, Slovakia | [відкрити контакти](див. вище в блоці «контактна інформація») | [відкрити контакти](див. вище в блоці «контактна інформація») | [відкрити контакти](див. вище в блоці «контактна інформація») | github.com/irapopovych

ABOUT

MSc student in Intelligent Systems at Technical University of Kosice, currently building a Knowledge Graph + LLM
agentic system for my thesis. Experienced across the full ML pipeline, from EDA and feature engineering to
model training and evaluation, with a focus on computer vision and medical image analysis. Working with LLMs
and agentic systems, with growing interest in NLP and large-scale data processing.

TECHNICAL SKILLS

Languages: Python, SQL
ML & Deep Learning: PyTorch, Scikit-learn, KMeans, MONAI, timm, NumPy, SciPy, Pandas, Matplotlib,
Seaborn
LLM & Agents: LLM APIs, Knowledge Graphs, GraphRAG, agentic pipelines, prompt engineering
Areas of Focus: Computer Vision, Medical Image Analysis, NLP, Large-Scale Data Processing,
Unsupervised Learning
Tools & Environment: Git, Docker, Linux, Jupyter, Google Colab, Kaggle, pytest

EDUCATION

MSc in Intelligent Systems (ongoing) Sep 2025 – present
Faculty of Electrical Engineering and Informatics, Technical University of Kosice

BSc in Intelligent Systems Sep 2022 – Jun 2025
Faculty of Electrical Engineering and Informatics, Technical University of Kosice

PROJECTS

Knowledge Graph + LLM Agentic System for Laboratory Data Analysis
MSc Thesis (ongoing)
Python · LLM APIs · Knowledge Graph · GraphRAG · Agentic AI · RAG
• Modeling heterogeneous lab data (device telemetry, maintenance logs, user records) as a Knowledge Graph
integrated with an LLM.
• Designing an agentic query pipeline that leverages graph-supported reasoning for decision-making tasks.
• Evaluating performance improvement of KG-augmented LLM vs. baseline LLM on practical query and
classification tasks.

Parkinson's Disease Diagnosis via Vision Transformers Sep 2024 – Jun 2025
Bachelor's Thesis
Python · PyTorch · MONAI · timm · 3D ViT · SegFormer3D · TimeSformer
• Classified Parkinson's disease from 3D DaTSCAN SPECT images using the PPMI database.
• Implemented and benchmarked 3D ViT, TimeSformer, and SegFormer3D for binary and 3-class supervised
classification.
• Best model (3D ViT with SegFormer3D pre-trained weights) achieved F1 = 0.8957 and accuracy = 0.8914.

Unsupervised Texture Segmentation via Classical CV Pipeline 2025
University Project
Python · Scikit-learn · KMeans · Gabor Filters · OpenCV
• Built a full unsupervised segmentation pipeline without deep learning: Gabor filter feature extraction, KMeans
clustering, morphological post-processing.
• Applied feature engineering (Gabor filters, local entropy maps) to capture texture complexity and patterns at a
local scale.
• Benchmarked on the Prague Texture Segmentation Benchmark; achieved Mean MS score of 49.24.

CNN Architecture Comparison: Inception-v3 vs. Inception-ResNet-v2 2026
Team University Project
Python · PyTorch · torchvision · Scikit-learn · Matplotlib
• Collaborated in a team to implement and compare two Inception-based architectures on image classification.
• Contributed to model training, evaluation pipeline, and analysis of accuracy, convergence, and compute trade-
offs.

ParkMe – Real-Time Parking App for Kosice 2025 – 2026
University Project in cooperation with FPT Software
Python · FastAPI · pytest · React Native · SQLAlchemy · Pydantic
• Part of a student team building a real-time parking application for the city of Kosice.
• Covered QA across the full stack: tested FastAPI backend (endpoints, authentication, business logic) and
React Native frontend.
• Wrote automated tests using pytest; performed manual testing and maintained test documentation.

ACTIVITIES

GymBeam Data Hackathon • Erste Digital Hackathon 2024 • T-Systems Hackathon 2024

LANGUAGES

Ukrainian — Native Slovak — B2 English — B1 Spanish — A1