I build complete, defensible data and AI systems in Python — from data acquisition, through models, to deployed and evaluated applications. Telecommunications student; open to AI Engineer, Data Scientist, and Junior Python / ML Engineer roles.
Every project below is defensible in a technical interview: each non-trivial decision is documented (ADRs), results are measured rather than asserted, and models are always compared against a baseline.
| Project | What it is | Live |
|---|---|---|
| ⭐ apply-scout | An LLM agent — a from-scratch tool loop with safety budgets and guardrails — that matches a job posting against a CV and GitHub evidence, with a trajectory-evaluation harness measuring success rate, citation fidelity, and cost per task. | demo |
| mlops-car-price | An MLOps layer around a price model: MLflow tracking and registry, a drift detector that is itself measured, and champion/challenger promotion decided by a paired bootstrap. | demo |
| ab-lab | A 5% test is only 5% if you look once, count each user once, and test one metric. Peeking, clustered users and many metrics each turn a true null into a 25–66% false positive rate — measured on experiments with no effect to find, each paired with the correction that puts it back. | demo |
| car-price-ml | The full ML cycle end-to-end: EDA, feature engineering, model comparison, SHAP, and a FastAPI + Docker prediction service. | demo |
- Languages — Python (primary), C++, SQL, JavaScript
- ML / Data Science — scikit-learn, LightGBM, pandas, SHAP, applied statistics, time-series forecasting
- LLM / AI — Anthropic API, from-scratch agents, QLoRA fine-tuning, evaluation harnesses
- MLOps — MLflow, drift monitoring, model registries, Docker, FastAPI, GitHub Actions
- NLP (Polish) — TF-IDF baselines, HerBERT / Bielik fine-tuning, PolEmo
| Project | Focus | Demo |
|---|---|---|
| apply-scout | LLM agent + evaluation | https://p0w3r223.github.io/apply-scout/ |
| mlops-car-price | MLOps / drift / promotion | https://p0w3r223.github.io/mlops-car-price/ |
| ab-lab | Applied statistics — when 5% is not 5% | https://p0w3r223.github.io/ab-lab/ |
| car-price-ml | Full ML cycle + API | https://p0w3r223.github.io/car-price-ml/ |
| wroclaw-air-insights | Data + 24h forecast | https://p0w3r223.github.io/wroclaw-air-insights/ |
| it-job-radar | Data engineering | https://p0w3r223.github.io/it-job-radar/ |
| pl-review-sense | Polish NLP sentiment | https://p0w3r223.github.io/pl-review-sense/ |
| pl-jobs-lora | QLoRA fine-tune (in progress) | https://p0w3r223.github.io/pl-jobs-lora/ |
- token-budget — a standard-library CLI that tracks token spend against a milestone budget.
- mini-traceroute — traceroute written from scratch in C++ over raw sockets.
- auth-log-scan — an OpenSSH auth-log scanner for brute-force and suspicious logins.