Дмитро

DMYTRO SHPAK
phone: [open contact info](look above in the "contact info" section) • Junior Data Engineer•
email: [open contact info](look above in the "contact info" section)
telegram : @dimaashcrypto
LINKEDIN PROFILE LINK

ABOUT MYSELF EDUCATION
Data professional with hands-on experience in fintech Qualified Worker Diploma in Electrical
and e-commerce. Worked as Data Engineer/Analyst Installation (Electrician)
on fraud detection projects at Monobank,
Bachelor’s degree System Analysis in
collaborating with ML engineers on data preparation,
TNTU, Ukraine
feature engineering, and EDA. Built price prediction
system for used goods marketplace — from data
cleaning to production integration.
1 year as Data Annotator validating ML models and
image datasets.
Tech stack: PostgreSQL, Python (pandas, numpy), Git
Turning raw data into actionable insights and ML-
ready features.

PROJECTS

2026 SKELAR X Monobank Monobazar price prediction API

Built an AI price prediction system for a used‑goods marketplace (SKELAR x Mono AI Competition 2026),
delivering 412ms response time and 82% pricing accuracy, and earned 3rd place in finals.
Designed semantic search with Supabase + pgvector and an embeddings pipeline (OpenAI
text‑embedding‐3‐small) to retrieve 20+ similar items.
Implemented similarity filtering and percentile‑based pricing (p20/p50/p80) plus ETL pipelines for market
data scraping (OLX, Rozetka, Prom).
Performed EDA on sold‑items data and engineered features like bargainMarkup and market metrics (sale
speed, price spread, demand).
Tech stack: PostgreSQL (pgvector), Python (pandas, numpy), Supabase, OpenAI API, Git.

2026 SKELAR X Monobank Fraud-detection project
Built an end‑to‑end fraud detection pipeline using BigQuery as the data source and Python for feature
engineering.
Engineered behavioral and transactional features (volume, diversity, velocity, error patterns, geo mismatch).
Trained a LightGBM classifier with cross‑validation and optimized the decision threshold for F1.
Achieved OOF F1 = 0.6527 with an optimal threshold of 0.49 and fraud rate 3.83% in training data.
Automated submission generation and optional upload to GCS for reproducible outputs.
Documented results and metrics for portfolio presentation.
GITHUB LINK

2024- Covizmo company, Siemens windturbine project
2025 Data annotation specialist
Annotated wind turbine technical condition data according to detailed guidelines for training a computer
vision model.
Worked within a corporate web‑based annotation platform to ensure secure, structured labeling workflows.
Used Postman API to upload, validate, and manage datasets and annotation tasks.
Tracked completed work and quality metrics in Google Sheets documentation.
Followed strict labeling standards and performed quality checks to keep annotations consistent and
reliable.
Documented edge cases, ambiguities, and labeling decisions to improve guideline clarity and model
performance.

2024 Udoma traffic aggency
Google ads/ facebook ads account farmer
Managed multi‑profile environments using anti‑detect browsers and proxy configurations to maintain
stable sessions and minimize risk signals.
Built and maintained structured account documentation (status, ownership, health checks, metadata) to
ensure traceability and auditability.
Performed cookie warm‑up and session preparation workflows to establish behavioral history and improve
account stability.
Set up billing profiles and payment bindings, validating readiness and minimizing verification issues.
Standardized SOPs for account lifecycle management, reducing errors and onboarding time for new
profiles.
Coordinated routine monitoring and reporting on account health, flags, and operational performance.