This gameplay video shows a Deep Reinforcement Learning agent playing Atari Breakout as it discovers and executes the "tunneling" exploit. Watch as the AI learns to strategically carve a narrow path through the bricks, trapping the ball at the top of the screen to rapidly accumulate points with minimal effort.
Try the app here

Dashboard that visualizes complex, multi-modal sensor data from dairy cows. High-dimensional features derived from wearable IMU movement tracking, spatial positioning, and physiological sensors are projected into an interactive 2D space to show behavioral clusters like feeding, drinking, and resting. Select any point to inspect the corresponding video frame, the cow's spatial location, and the specific sensor signals driving the model's predictions.
Try the app here

This interactive environment visualizes relationships between words and concepts by using high-dimensional embedding data reduced to 3D space. In this model, distance represents meaning, so words that are conceptually similar naturally gravitate toward one another to form distinct, intuitive clusters.
Try the app here

An interactive web application for visualizing how a Convolutional Neural Network (CNN) perceives clothing images from the Fashion-MNIST dataset. The app projects high-dimensional CNN embeddings into 3D using t-SNE, hovering over points reveals the original image, and the visualization helps illustrate how deep learning models organize visual information in feature space.
Try the app here

An interactive web application that visualizes how a pre-trained Convolutional Neural Network processes handwritten digits in real-time. Users can draw digits on a 28×28 grid and watch as their input flows through the network layers in 3D visualization. The app shows convolutional filters, activation maps, pooling operations, fully connected layers with live connection strengths and neuron activations, and confidence predictions for each digit class.
Try the app here

Battery Evaluation And Classification Of Next-gen Vehicles — made with the Antemia GmbH team. Users can adjust cell types (Prismatic, Cylindrical, Pouch), capacities, voltages, and module counts, with real-time updates to energy calculations and 3D visualizations. The app uses Principal Component Analysis to compare the current EV configuration with market segment clusters based on performance and size metrics.
Try the app here

Web app for creating a custom track via interactive splines, it also computes an optimal racing line by dividing the track into sectors and refining positions in each sector with CMA-ES (Covariance Matrix Adaptation Evolution Strategy). The lap time estimation uses a double pass speed method: the first pass handles acceleration constraints, while the second pass enforces deceleration limits, producing a speed profile. Users can compare multiple runs with varying parameters to see how these changes affect lap times and optimal racing lines.
Try the app here

A Vue.js web application that provides an intuitive interface for creating and managing tabular data. Users can dynamically add or remove rows and columns, edit cell contents including image insertion, and get basic statistical insights for numeric columns. The application features export capabilities for JSON and PDF formats, multilingual support with a language switcher, and a responsive design that works across all devices.
Try the app here