A participant at Recurse Center, a Brooklyn-based programming retreat, detailed a range of technical projects completed during a summer residency. The individual joined study groups including Agentic Adventures, which examined large language models and agent systems through sandboxed experiments with tools like Ollama and Docker. One project involved building a cooperative game framework to test agent collaboration, while another used minGPT to create a Shakespearean text predictor trained on Romeo and Juliet.

Practical Deep Learning sessions followed the Practical Deep Learning textbook by Jeremy Howard and Sylvain Gugger, focusing on hands-on implementation of classical machine learning methods and neural network construction. Mathematical exploration occurred through Math Monday gatherings, where participants solved Project Euler problems, analyzed fractal generation, and developed Voronoi diagram applications, including a Hilbert curve generator for pen plotter output.

Collaborative work included reverse-engineering the Keldon AI for Race for the Galaxy, revealing strategic biases toward economic over military tactics in the game's neural network weights. A separate initiative implemented DEFLATE compression, combining LZ77 backreferences and Huffman coding, using Rust to build a decompressor from specification. Pairing sessions also covered SAT solver applications for Sudoku, shell-based Game of Life visualizations, and Magit interface development in Emacs.

Experiments extended to LLM-driven game design, including a word-guessing variant where agent "personalities" influenced hint generation, and a diplomacy simulation testing cooperative behavior. A pronunciation-matching tree search connected Indonesian and Farsi vocabulary through phonetic similarity, while a constructed language project drew on PHOIBLE data and APiCS linguistic patterns to create cross-cultural lexical foundations. The participant also developed a WebLLM integration component to enable user-hosted GPU processing for shared project access.

Physical computing activities included 3D printing a birthday candle holder and generating Go game snapshots via the hub's pen plotter. Weekly board game sessions covered Go instruction and Gowanus club visits. Other projects featured biometric aura visualization, noise-reactive decorations, e-ink screen game prototypes, and receipt printer-based Exquisite Corpse games.

The residency concluded with a presentation of completed work, supported by AI-assisted slide preparation. A final reflection documented the experience, emphasizing iterative learning and cross-disciplinary experimentation within the program's collaborative environment.