MSc InfoBioChem student at Gdańsk University of Technology, building reproducible cheminformatics and data-analysis pipelines, from molecular descriptors and QSPR models to LC-MS lipidomics. I care about validation: what a model can and cannot predict, and whether a result reproduces from the data.
- Core: Python · pandas / NumPy / SciPy · scikit-learn · RDKit · SQL
- Scientific: chemometrics · QSAR/QSPR · cheminformatics · omics statistics (LC-MS lipidomics) · mechanistic modelling
- Other: MATLAB · SAS · PostgreSQL · pytest and GitHub Actions
| Project | Problem → result | Tech |
|---|---|---|
| qspr-microplastic-sorption | Predicting pollutant sorption on PP microplastics from 35 compounds. High-Q² models were rejected for chance correlation; a 2-descriptor MLR was stable over 200 splits. One command reproduces the report, with leakage tests in CI. | Python, scikit-learn, PaDEL, MOPAC |
| lipidomics-human-milk-untargeted | Untargeted LC-MS of human milk, month 1 vs 12. QC-driven filtering (752 → 94 features) and donor-level paired statistics, with restricted data handled through pseudonymisation and synthetic-data tests. | Python, SciPy, scikit-learn |
| cheminformatics-chemical-space | Does molecular representation change what "similar" means? Physchem vs MACCS vs Morgan on 90 fragrance molecules; odour class is only weakly encoded by structure. | RDKit, PubChem, PCA / t-SNE / UMAP |
| moodle-sync | A self-hosted tool that syncs Moodle files, deadlines and grades to the cloud, a calendar and notifications. Cross-platform CI, secret masking in logs, a documented data-retention table. | Python, REST APIs, GitHub Actions |
- biochemical-process-modeling: kinetics, CSTR and Monod chemostat fits with bootstrap CIs; spectral pKa.
- statistics-coursework: genetic association analysis in SAS; GC-MS signal processing in MATLAB.
- open-databases-structural-analysis: crystal structure of a cisplatin analogue (CSD, Hirshfeld surfaces); PCR primer design.
- postgresql-database-apps: normalised PostgreSQL schema with a least-privilege role.
- delhi-climate-analysis: pandas data cleaning and seasonality.
- amp-cubecoders-tg-bot: Telegram bot for AMP game servers, with checksum-verified self-update.
- family-pet-bot: an AI "pet" for family chats, with a clear statement of what goes to the model provider.
AI-assisted development was used for implementation and documentation. Method choice, validation strategy, data-handling decisions, result verification and interpretation were reviewed and owned by me. Each repository says this in its README.
🇵🇱 Student studiów II stopnia na kierunku InfoBioChem na Politechnice Gdańskiej. Buduję odtwarzalne pipeline'y chemoinformatyczne i analityczne. Powyżej są projekty studenckie i poboczne; projekty studenckie mają krótką sekcję po polsku.