Lab & Projects
AI/ML and security projects I've built while learning cybersecurity. From traditional security tools to machine learning-based threat detection—all real, working implementations with code and documentation.
Cybersecurity Projects
Sentinel-SSH
Security Tooling
SSH intrusion detection system I built to learn security development. Real-time journald log parsing, brute-force detection, and automatic UFW blocking. My first production security tool.
Key Features:
- Real-time journald log parsing
- Smart brute-force detection (sliding windows)
- Temporary UFW blocks (auto-expire to prevent lockout)
- Prometheus metrics integration
- Documented extensively (learning by writing)
Stack:
Threat Model
Protects against SSH brute-force, credential stuffing, and automated scanners
RL-IDS: Reinforcement Learning Intrusion Detection
AI/ML Security
Adaptive intrusion detection system using Deep Q-Network that achieved 93% detection confidence with sub-100ms latency. Built a modular Python framework with FastAPI, PyTorch, and Scapy for live packet capture and real-time inference.
Key Features:
- Deep Q-Network for adaptive learning
- Live packet capture and analysis
- Real-time inference engine
- Outperforms traditional rule-based systems
- Modular architecture with FastAPI
Stack:
Threat Model
Machine learning approach to detect novel attack patterns and zero-day exploits
Redactify - PII Detection & Redaction
AI/ML Security
CNN-based deep learning model to detect and redact PII (Personally Identifiable Information) from government documents with 96% accuracy. Uses YOLOv7 for detection and Hugging Face transformers for redaction.
Key Features:
- YOLOv7 for image-based PII detection
- OCR for text extraction
- Hugging Face transformers for NLP redaction
- User-friendly Gradio interface
- Government document compliance
Stack:
Threat Model
Protects sensitive personally identifiable information from unauthorized disclosure
AI/ML Projects
Water Potability Detection
MLOps
End-to-end MLOps pipeline for predicting water potability based on quality parameters. Features CI/CD automation, Docker containerization, FastAPI deployment, and monitoring infrastructure.
Key Features:
- Complete MLOps pipeline (train → deploy → monitor)
- FastAPI web interface
- Docker containerization
- CI/CD automation
- Model monitoring and versioning
Stack:
Threat Model
Data validation and model integrity checks to prevent poisoning attacks
Credit Card Fraud Detection
Machine Learning
Machine learning system for detecting fraudulent credit card transactions using ensemble methods. Built Flask RESTful API for real-time fraud detection with high accuracy.
Key Features:
- Supervised learning with ensemble methods
- Flask RESTful API for predictions
- Real-time fraud detection
- Data visualization dashboard
- High accuracy on imbalanced datasets
Stack:
Threat Model
Detects fraudulent transaction patterns and anomalies in real-time
Learning Lab
Homelab Infrastructure
Learning Lab
Personal lab for learning infrastructure and security. Running on Raspberry Pi hardware with Pi-hole DNS filtering, SSH monitoring, and basic observability. Still learning and breaking things regularly.
Key Features:
- Pi-hole for DNS-level ad/tracker blocking
- SSH hardening and monitoring practice
- Prometheus + Grafana for observability learning
- Network topology experiments (sometimes RFC-violating)
- Safe environment to break things
Stack:
Threat Model
Learning lab - experiments with security configurations and monitoring
Known Issues & Limitations
Sentinel-SSH: May block your own IP if you mistype your password multiple times. Whitelist logic exists, use it.
RL-IDS: Reinforcement learning model needs continuous retraining on new attack patterns. 93% accuracy is good but not perfect.
Homelab: Still learning and experimenting. Things break regularly, which is exactly the point of having a lab.
Reality check: These are learning projects built by a student. They work for my use cases and have been tested, but they're not battle-tested enterprise solutions. Use at your own risk and don't blame me if things go sideways.
Want to Learn Together?
Found a bug? Have feedback? Want to collaborate on security projects? I'm always looking to learn from other people building security tools. Open an issue or reach out.