💎 Non ad originem, sed ad qualitatem (Don't judge by the origin, judge by its quality)

🧑‍⚖️ Quis es tu, ut te iudicem tribunal mundi? (Who are you to appoint yourself judge of the world?)

Student focused on computer science, mathematics, and systems engineering, with hands-on experience developing software, managing Linux systems, analyzing network traffic, and building machine learning pipelines.

I work across the stack, from system configuration and network analysis to application development, data analysis, and model evaluation. My projects emphasize practical engineering, measurable results, reliability, and a clear understanding of system behavior.

• Linux systems administration and service management
• Network traffic analysis and intrusion detection
• Software development and system architecture
• Applied machine learning and statistical evaluation
• Data analysis, model robustness, and interpretability

Languages: Python (proficient), C and SQL (basics)

Systems & Networking: Linux, Arch Linux, systemd, tshark, TCP/IP

Data Science & ML: NumPy, pandas, scikit-learn, PyTorch, SHAP, Matplotlib

Development Tools: Git, Jupyter, VS Code and derivatives

Peer-to-peer messaging application providing end-to-end encryption, authenticated contacts, and protection against passive network observation and active man-in-the-middle attacks.

• Implementation: Python-based CLI/TUI application using X25519, HKDF, AES-GCM, Argon2id, and SQLCipher.
• Security Features: Encrypted identity vault, fixed-size 4096-byte packet padding, out-of-band fingerprint verification, and hardened storage and memory handling.
• Security Model: Documented threat model, clearly defined security assumptions, and explicit security limitations.
• Status: Stable release. The roadmap includes Tor/Onion integration, enhanced post-compromise security, and migration of security-critical components to memory-safe languages.
• Packaging: Cross-distribution Linux AppImage and standalone Windows executable.

Lightweight, heuristic-based Intrusion Detection System for real-time network traffic analysis and anomaly detection in home labs.

• Implementation: Python/Flask backend, tshark packet capture, heuristic scoring engine, JavaScript/Chart.js dashboard, and SQLite persistence.
• Features: Configurable thresholds, graylist/watchlist/whitelist, authentication, and real-time traffic monitoring.
• Deployment: Tested on a Raspberry Pi 5 in a home network with approximately 10–12 devices and a 20 Mbps connection.
• Resource usage: Under 1% CPU direct at idle and approximately 220–330 MB RAM.
• Detection capabilities: Port scans, flooding/DoS patterns, and suspicious service probing.
• Status: v3.0.0.

Free, open-source, self-hosted flashcard platform combining rating-based study workflows with Anki-style openness, without mandatory spaced repetition.

• Implementation: Minimal Python HTTP server using only the standard library, with a vanilla JavaScript frontend.
• Storage: Filesystem-based CSV decks and persistent JSON ratings.
• Features: REST API, HTML sanitization, import/export, editable decks, filtering by rating/tag/subdeck, and mobile-friendly UI.
• Deployment: Self-hosted, device-independent, and compatible with systemd.
• Status: v1.0.0.

• Model: Triple-scenario model engine using Random Forest and Logistic Regression.
• Task: Predict credit card default from structured financial and behavioral data.
• Evaluation: Positive mean net profit on the test set based on backtested model predictions.
• Focus: Business-driven evaluation under real-world hypotheses and noisy scenarios.

• Model: Random Forest.
• Task: Predict next-day rain from historical meteorological data.
• Results: Accuracy 71.35%, ROC-AUC 77.27%.
• Focus: Classification using real-world meteorological data with noisy features.

• Model: Logistic Regression.
• Task: Predict stress type from a 25-question survey.
• Results: Macro F1 = 0.8538.
• Focus: Multiclass classification under significant class imbalance.

• Model Robustness & Stress Testing: Analyzing ensemble performance degradation under progressive feature omission and noise injection.
• Explainability & Decision Limits: Investigating global and local interpretability in tree-based models, including SHAP, and evaluating threshold sensitivity under asymmetric loss.
• Production-Oriented ML Engineering: Developing vectorized validation pipelines, probability calibration, and empirical performance evaluation using frequentist confidence intervals.

• GitHub: FreyFlyy
• LinkedIn: Francesco Scolz
• Kaggle: freyfly
• Hugging Face: FreyFlyy