Tocsin
​​
2025
​​​
​
​
For this project, I built a custom Large Language Model (LLM) using TensorFlow, which is trained specifically on the Enron Corpus. This model is not a generic tool but a targeted instrument, designed to generate stochastic, karaoke-style lyrics from the raw, unfiltered language of corporate communications. The goal is to create a disembodied experience: the familiar, playful format of karaoke is subverted by lyrics sourced from the language of a financial felony. The work then layers this text with hand-drawn video, which combines my own footage with imagery drawn from the LAION dataset, a massive collection used to train AI image generation models.
Tocsin is therefore a collision of analog and digital, human and machine. It is a system built to reveal a system, a tool for both artistic expression and critical investigation. By inviting the audience to perform these lyrics, the project implicates them in the very absurdity and opacity it seeks to expose.
​
​