I believe the best way to understand anything is by diving in and experimenting. With that mindset, I’ve embarked on a series of hands-on projects to explore what artificial intelligence can do. Here, you’ll find a selection of AI experiments I’ve undertaken, along with the insights and discoveries they revealed about the potential of this transformative technology.
| Model | Date | Experiment | Demo Link |
| ChatGPT4.0 | Jan 2025 | Idea/Objective: Create a javascript version of hangman, however I asked it to replace the hangman imagery with a sci-fi theme. ChatGPT picked the theme of an alien invasion. Notes: The coding took five iterations to produce a functional result, with most issues centered around an unusable UI. The workaround was to make the prompt extremely specific about expected behavior. For example, I included instructions such as: “The screen should update dynamically as the user plays to indicate which letters have been guessed previously.” The graphical representation of the alien invasion missed the mark somewhat, but I found it funny and chose not to ask the AI to correct it. | Demo |
| ChatGPT3.5 & 4.0 | Feb 2025 | Idea/Objective: Create a music theory reference website that lets users quickly calculate first-inversion chord spellings for common progressions (e.g., I–IV–V) across different keys (e.g., A, C, G♯) and modes/scales (e.g., Major, Minor, Dorian). Notes: Using ChatGPT 3.5, it took about 15 iterations to generate a functional interface for the tool, but the chord calculations were often inaccurate. Switching to ChatGPT 4.0 produced a much more reliable reference page in only four iterations. The only manual fix required was adjusting the alignment of the modal windows. In the GPT-generated version, the piano keys were also spaced awkwardly. | Demo |
| ChatGPT 4.0 | Feb 2025 | Idea/Objective: Rapidly prototype a Eurorack module concept by using ChatGPT 4.0 to build an interactive web demo demonstrating subtractive synthesis. Notes: After discussing a new module concept with a Eurorack manufacturer, I explored whether ChatGPT could quickly generate a working prototype. After several failed attempts, a detailed prompt requesting a single-file site (HTML, CSS, JavaScript, and the Web Audio API) produced a successful result. The tool included an 8-step sequencer with pitch and velocity controls, three configurable oscillators, a tempo control, and transport functions (start, stop, randomize, reset) along with a help modal. The page, titled “Simple Euphonica Demo,” used labeled controls and a clean, color-coded layout to organize each section. | Demo |
| ChatGPT 4.1 | April 2025 | Idea/Objective: Create an interactive audiovisual grid using only HTML, CSS, and JavaScript, where squares change color and generate sound as the user moves the cursor across them. Notes: Using ChatGPT 4.1, I iteratively developed a website featuring a square grid composed of 50 smaller squares. Each square begins white with a black border and changes color when hovered over, while adjacent squares shift to lighter shades. I refined the behavior to include faster fade-ins (3 ms), slower fade-outs (2 seconds), additional color gradients for neighboring squares, and a reset feature that returns all squares to white if the cursor remains idle for four seconds. I also added a set of color-selection buttons that allow the user to change the active color scheme, with blue as the default on page load. Further iterations introduced sound generation using the Web Audio API. Each square triggers a tone when the cursor enters it, with pitch determined by its position in the grid and timbre determined by the selected color (e.g., bell for blue, beep for green, buzz for red). Sounds fade out over one second and continue playing even if the cursor leaves the square. During development, I encountered an issue where sound would not play when the page first loaded. ChatGPT explained that modern browsers block audio playback until the user performs a gesture (such as a click or keypress), meaning the sound cannot automatically start on page load without user interaction | Demo |
| ChatGPT 4.1 | April 2025 | Idea/Objective: Create a single-file website (HTML, CSS, and JavaScript) that visually demonstrates Euclidean rhythms with multiple percussion layers. Notes: Using ChatGPT 4.1, I generated a tool with four rhythmic layers—kick, snare, tom, and hi-hat—each capable of producing Euclidean patterns. Users can adjust the number of divisions for each layer while a master BPM control governs the overall tempo. Start and stop controls allow the sequence to be played or paused. In later iterations, I replaced the per-layer BPM controls with a rotation setting, allowing users to offset the starting position of each Euclidean pattern. I also added a randomizer button next to the start control, which randomizes all layer settings while preserving the master BPM. | Demo |
| ChatGPT 4.1 | May 2025 | Idea/Objective: Create an interactive website version of Andrew Huang’s Book of Chances card deck. Notes: Using ChatGPT 4.1, I first asked whether it was familiar with Andrew Huang’s Book of Chances card deck and whether it could recreate it as an interactive website. The initial version was immediately usable, but I spent about an hour iterating on improvements. I also prompted the AI for suggestions, which led to additional features such as the ability to reorder the cards and copy the currently selected cards to the clipboard. | Demo |
| ChatGPT 4.1 | June 2025 | Idea/Objective: Create a simple browser-based music tracker with the ability to save patterns. Notes: Using ChatGPT 4.1, I asked it to generate a basic music tracker application. In the initial version, the ADSR envelope was implemented as a global control using sliders. When I requested that ADSR be incorporated into each individual track instead, the AI independently chose to replace the slider controls with descriptive envelope presets. This turned out to be an effective design choice and improved the usability of the interface. | Demo |
| ChatGPT + veo 3 | Dec 2025 | Idea/Objective: Experiment with the possibilities of using AI to create a short film. Notes: Details captured here: Link | Demo |
| ChatGPT 5.2 | Jan 2026 | Idea/Objective: Create a tool that generates short, randomized piano practice routines (10–15 minutes) focused on specific skill development. Notes: Using ChatGPT 5.2, I asked it to build a tool that would help me practice piano more consistently by generating focused, randomized practice routines. The exercises targeted three areas: improving familiarity with key signatures and modes, building a library of rhythmic and comping patterns, and developing fluency with chord inversions—especially in less familiar keys. I provided detailed guidance on the types of exercises the generator should produce. After extensive iteration—about 15 versions—the tool became quite usable. Improvements included features such as a one-shot “listen” button for chords and an expanded range of available time signatures. After using the tool for several days, I requested an additional refinement: the randomization system should default to common Western conventions (major/minor modes and 4/4 or 3/4 time) while still allowing manual selection of more advanced modes and time signatures when desired. | Demo |