I am a Python developer focused on automation, local AI inference, defensive security and software that interacts with real hardware.
My flagship project is CatchBot, a modular asynchronous automation platform with a CAPTCHA model that I trained and deployed myself. I am also building Security Lab, a collection of transparent defensive security tools, and Joy-Pi Casino, a complete Raspberry Pi hardware project.
CatchBot is my most advanced project. It combines a modular event-driven Python automation system with a complete machine-learning workflow for solving image CAPTCHAs locally. I fine-tuned a TrOCR-based vision encoder-decoder on more than 5,000 labeled images, converted it into separate ONNX encoder and decoder runtimes, and connected it to the application's asynchronous processing pipeline.
Key engineering highlights:
- More than 5,000 labeled training images and a model with roughly 334 million parameters
- Custom data collection, augmentation, training and exact-sequence evaluation pipeline
- Approximately 98% CAPTCHA accuracy in the project's evaluation
- Local ONNX Runtime inference on CPU or CUDA with no paid API dependency
- Test-time augmentation, confidence handling and prediction feedback
- End-to-end image detection, preprocessing, inference and answer submission
- Roughly 10,000 lines of Python across 38 modules and more than 20 composable mixins
- Dedicated packages for core logic, configuration, CAPTCHA handling, features, UI and webhooks
- Multi-account process orchestration, proxy support, rate-limit handling, persistent configuration, logging and statistics
View CatchBot source | Download the latest release
Security Lab is a project-oriented monorepo for small, auditable defensive security tools. Its first working project is the Security Log Analyzer, a local command-line application for finding suspicious behavior in server and authentication logs.
- Parses Apache, Nginx and OpenSSH authentication logs
- Detects scanning, repeated login failures, password spraying, distributed failures and successful logins after repeated failures
- Follows growing log files with bounded retained detection state and common rotation detection
- Produces Rich terminal, JSON and CSV reports
- Uses configurable thresholds, CIDR allowlists and strict validation
- Runs automated tests, linting, type checks, security analysis and dependency auditing in CI
View Security Lab | Open the analyzer documentation
Joy-Pi Casino is a complete interactive casino terminal running on real Raspberry Pi hardware.
- RFID authentication with persistent SQLite wallets
- Five playable games with separate rules engines and display layers
- Physical buttons, an LED matrix, a character LCD and sound feedback
- Modular hardware drivers and game logic that can also be tested on a normal PC
| Area | Technologies |
|---|---|
| Development | Python, Java, JavaScript, PHP |
| AI and data | PyTorch, Hugging Face Transformers, TrOCR, ONNX Runtime, NumPy, Pillow, JSON, CSV, SQLite, MySQL |
| Systems | Linux, Windows, Docker, networking, Raspberry Pi |
| Tooling | Git, GitHub Actions, Bash, PowerShell, VS Code |
You can find more of my work on GitHub, visit mynameiskillua.de, or contact me directly on Discord.
