LLM

The dataset release for the ChatTLA+ paper — SFT corpus and benchmark for TLA+ spec generation, posted anonymized for blind review.

Two homeworks from Loyola’s special-topics LLMs class: prompting GPT-2 and comparing base vs instruction-tuned Qwen.

A voice-first RAG chatbot trained on Loyola’s ITS knowledge base, built after two years of answering the same tickets.

Fine tuning GPT-2 to talk like Yeat

Early exploratory repo for generating TLA+ specs from natural-language requirements with an LLM, with a TLC harness wired in. The scaffold that eventually grew into ChatTLA+.

A toy transformer that reads pyfiglet ASCII art and tries to spit the original word back out. Calling it an LLM is generous.

A toy where two local Ollama models loop on consciousness until one of them gives up

Obfuscator for public Python code to confuse LLMs

A VS Code extension I made when I kept running out of Copilot credits — flattens a repo into one .txt file you can paste into a free LLM.

A small Python CLI for chatting with OpenAI and running English-to-Bash commands from the terminal. The thing I actually open when I want to talk to a model.