Projects
Built andreleased.
Everything here is public on GitHub: machine learning, computer vision, data pipelines, plus a few things running on hardware.
B.Eng. thesis: predicts which parts of YouTube videos viewers replay the most. Frozen video, audio and subtitle encoders, modality fusion and a temporal model, trained on the Mr.HiSum dataset (over 30,000 videos).
Vehicle detection and recognition on road footage with a fine-tuned YOLOv8 model, covering dataset preparation and inference on video.
End-to-end ETL into a star-schema data warehouse for online store data (over 100,000 orders), with a Streamlit KPI dashboard. Fully containerised with Docker.
Credit default prediction on the Kaggle GiveMeSomeCredit dataset: data cleaning, feature engineering, model training and evaluation.
Detects safety helmets and bare heads in images and video with a fine-tuned YOLOv8 - a starting point for construction site safety monitoring.
A scheduled ETL pipeline: pulls weather for six Polish cities from OpenWeatherMap, cleans it up and stores it in PostgreSQL.
Work in progress. A real-time pipeline meant to catch divergence between crypto prices and the sentiment of news headlines. Sentiment analysis with FinBERT and the database schema are done; Kafka streams and the dashboard are under way.
Web application generating trading signals from market data pulled over a REST API, with a Java backend and a browser front end.
Generates printable posters from a Spotify playlist: pulls track data and cover art through the Spotify API and lays it out automatically.
Multi-threaded scraper collecting job offers from listing sites with Selenium, built to keep throughput up across many pages at once.
An ESP32 device that polls a list of websites over Wi-Fi and reports their availability - an uptime monitor running on a microcontroller.