Neural network-based application designed to predict the behavior of damped vibrations in single degree of freedom mechanical systems. Using supervised learning, the model is trained on solutions to differential equations of motion. The app features real-time visualization of the learning process and allows users to input their own parameters to test the neural network’s predictive capabilities. This combination of AI and dynamic simulation makes app a powerful educational and research tool for understanding vibration analysis.

This project is a comprehensive full-stack engineering application featuring a robust FastAPI backend that serves a dual-purpose RESTful API. The core functionality is an analytical engine that calculates support reactions, shear forces, and bending moments for statically determinate beams, processing both structured JSON inputs and unstructured natural language descriptions by integrating with the OpenRouter AI API (GPT-4o) for advanced text parsing. The results are delivered to a clean, responsive frontend built with vanilla JavaScript and Chart.js, which renders interactive and precise diagrams, demonstrating a complete pipeline from user input to complex calculation and data visualization.

An experimental multiplayer web app where players compete to train neural networks to solve a ballistic problem. Each user controls their own AI agent, adjusting weights in real time. A shooting simulation visualized with p5.js and Matter.js shows how well the AI performs. The faster and more accurate the AI learns, the better the player's score. A blend of real-time training, physics, and fun.

This is a simple digit recognition app that uses a neural network trained on the MNIST dataset. Users can draw digits (0-9) on a canvas, and the model will try to predict the number in real time. The app demonstrates how neural networks can process visual input directly from user handwriting. Note: Due to Google security constraints (for example, cross-origin restrictions), the model does not work online and must be run locally. To run locally: See the README file in the GitHub repository for setup instructions.

AI-powered O-ring detection system with automatic diameter measurement. The model was trained locally using a dataset from Roboflow in PyTorch and converted to ONNX (Open Neural Network Exchange) format for efficient deployment. After calibration with a known object, the system can accurately measure the diameter of O-rings. The application uses React and WebGPU for fast in-browser inference and visualization. Due to the prototype nature of this project, please read the README for current limitations of the application.

This AI Snake game implements a Q-learning algorithm to train a snake. The project features real-time visualization using SDL, dynamic exploration rates for balanced learning, and performance tracking with charts. Key optimizations include penalties for the snake getting too close to its body, and a reward system that encourages efficient pathfinding. The snake learns from both positive rewards (food collection) and negative penalties (collisions, circling behavior), with the AI adapting its strategy over thousands of episodes. The code supports native execution (with SDL).

Job Application Agent is a Node.js application that automates the process of applying for jobs by searching listings via the Jooble API, analyzing job descriptions, and using AI (Ollama) to personalize your CV for each role—while preserving your original CV format and supporting both English and Polish markets.