Nine applications built by members and running on the club’s own servers, from computer vision and language models to classic algorithms. Every project links to its source code.
A convolutional network trained on MNIST classifies handwritten digits as they are drawn and displays the full probability distribution. Strokes are normalized with the same procedure used to build the dataset. Test accuracy is 98.6 percent.
PyTorch · CNN · Canvas API · MNIST
Play Stockfish 16 at adjustable strength from 100 to 3600 ELO with a live evaluation bar, captured piece display, algebraic move history, and a hint engine, driven by a python-chess backend.
Python · Stockfish 16 · UCI Protocol · chess.js
A*, Dijkstra, Greedy Best-First, BFS, and DFS animated on an editable grid with walls, weighted terrain, and generated mazes. Each run reports nodes visited, path cost, and whether the algorithm guarantees a shortest path.
JavaScript · A* Search · Binary Heap · Maze Generation
Candlestick charts, volume, and company statistics for any ticker over selectable periods, with recent headlines classified by a DistilRoBERTa transformer fine-tuned on financial news and aggregated per stock.
Transformers · DistilRoBERTa · yfinance · Plotly
Classifies images with ResNet-50, EfficientNet-B0, or MobileNet-V3 pretrained on ImageNet. Inference runs on the club GPU when available and falls back to CPU automatically.
PyTorch · ResNet-50 · ImageNet · GPU/CPU Failover
Generates a short video clip from a still image and a written prompt using a video diffusion model. Includes an ethics agreement, consent rules for images of real people, and a guide to recognizing synthetic video.
Motion Models · Club GPUs · AI Ethics
Restyles photos with an image-to-image diffusion model driven by a written prompt, with adjustable strength. Six filters built from classical image-processing operations are available as a separate mode.
PIL · NumPy · Convolution Kernels
Configure a network architecture, dataset, and optimizer, then follow the decision boundary and training metrics as they stream from a PyTorch training loop on the server.
PyTorch · SSE Streaming · Plotly · Flask
A character-level n-gram model built from classic texts or any pasted sample. Context length and temperature sliders control coherence and variety, the same parameters used in modern language models.
Python · N-gram Models · Temperature Sampling
