Anthony Zatika

I built a modular machine-learning intrusion-detection system that was presented at the 2024 IEEE/ACM International Conference on Utility and Cloud Computing. The system normalizes network security datasets such as CICIDS2017, UNSW-NB15, and ToN-IoT for a reproducible IDS workflow, then trains and evaluates models including XGBoost, Random Forest, SVM, and neural network baselines. I designed the end-to-end pipeline for preprocessing, label encoding, hyperparameter optimization with Optuna, model export, and real-time Zeek conn.log analysis. The runtime monitor extracts engineered traffic features, scores live network activity, and supports optional IP reputation enrichment for operational context. My role encompassed system architecture, data engineering, model selection and evaluation, reproducibility, and production-oriented hardening.

I designed and built an autonomous cyber-defense agent alliance presented at AICCC 2025 in Tokyo, Japan. The system coordinates lightweight artifact capture, intelligent routing, specialized detection agents, audit logging, confidence thresholding, and human-review controls. A routing layer triages artifacts into network traffic, executable and malware analysis, or rejection paths, then dispatches them to specialized modules, including an XGBoost-style network IDS agent, PE static-analysis agent, and opcode-sequence HMM malware-analysis agent. I implemented the full orchestration layer, including decision policies, risk scoring, feedback capture, JSONL audit trails, CLI tooling, CI tests, and extension points for PPO-based routing and production model artifacts. The project demonstrates safe autonomy for cyber defense while preserving traceability, measurable risk scoring, and analyst oversight.