Віктор

Viktor Luka [відкрити контакти](див. вище в блоці «контактна інформація») [відкрити контакти](див. вище в блоці «контактна інформація»)
Lviv, Ukraine Open to work: Remote
Data Analyst LinkedIn Telegram GitHub

ABOUT ME
Developer transitioning into Data Science, with a web background (Shopify, React/Next.js, NestJS) and
hands-on experience with AI solutions (Claude API, MCP, vector search). I moved from web development to
data science to apply my mathematical training and build end-to-end ML solutions in Python (Pandas,
PyTorch, scikit-learn, SQL). Team-oriented, I automate routine work and keep learning through
university, courses and AI/data events.

WORK EXPERIENCE
Junior Full Stack Developer Dec 2025 - Jun 2026
KodiX
Built and delivered the frontend of client web applications with React and Next.js
Developed REST APIs with NestJS and an ORM; worked in a team with Git, code review and CI
Learned how to use Claude Code skills, hooks, plugins and slash commands for automation tasks
Learned AI-agent engineering (MCP, Claude API, vector search)

Trainee Shopify Developer Jul 2025 - Oct 2025
Hallwil
Customized e-commerce storefronts and integrated store data through Shopify APIs
Worked in the Shopify admin and on the storefront frontend to improve business
Collaborated in a team: solved issues, took part in code reviews and workflow improvements

EDUCATION
Bachelor in Computational Intelligence of Smart Systems 2023 - 2027
Lviv Polytechnic National University
Course “Data Science Essentials with Python” Dec 2025
Cisco Networking Academy
Winter School “A European approach to Data Governance” Feb 2026
Data4EU, Jean Monnet Module · Lviv Polytechnic National University
Summer School “Regulating AI and Data in the Age of EU Digital Reforms” Aug 2026
Data4EU, Jean Monnet Module · Lviv Polytechnic National University

PET PROJECTS
Customer Churn Prediction Demand Forecasting
Business problem: decide which telecom customers get a Business problem: decide how much to order per store and
costly retention offer, so the budget goes to those worth product for the next 28 days, to keep order cost down.
it, not to everyone.
Data: IBM Telco, 7,043 customers, 26.5% churn (about 1 in 4 Data: Favorita grocery sales: 25 store-family series over
leaves); leakage columns removed so scores are not 731 days (about 2 years); holidays and missing days handled.
inflated.
Approach: compared interpretable Logistic Regression with Approach: baseline vs XGBoost, LightGBM, RNN and LSTM; walk-
Random Forest, XGBoost and a PyTorch MLP; 5-fold cross- forward validation to avoid look-ahead leakage; data queried
validation; contact threshold chosen by expected profit, with SQL in PostgreSQL.
not accuracy.
Results: all models reached ≈0.86 ROC-AUC, so the simple, Results: XGBoost WAPE 0.105 vs 0.141 for the baseline (about
explainable Logistic Regression was kept; simulated profit 25% lower error); simulated order cost -23%; LSTM was no
+9.5% vs contacting all customers. better and 50x slower.
Takeaway: a business-cost threshold matters more than model Takeaway: tree models on a scaled target beat neural nets
complexity; designed an A/B test to validate the offer on here, and faster; next step: quantile forecasts.
real customers.

LANGUAGES TECHNICAL SKILLS
English (Intermediate) Python & Data Analysis Mathematics
Ukrainian (Native) NumPy, Pandas Linear Algebra, Calculus
Matplotlib, Seaborn Probability, Statistics
SOFT SKILLS Jupyter, EDA, Data Cleaning Hypothesis Testing, A/B testing
Responsibility Scikit-Learn, XGBoost
Initiative
Adaptability Databases & Tools BI & Spreadsheets
Curiosity SQL, PostgreSQL Excel, Google Sheet
Time management Git, Docker, CI Power BI, Looker Studio
Attention to detail
Work ethic