Fashionably Late

DMSB

Class of 2029

Business Administration: Finance, AI Minor

Artist Statement

My game is called Fashionably Late (782 lines). First, you will receive a prompt of an occasion that you need an outfit for. Then, you need to pick the clothing items and accessories that would work for the occasion (you will switch lanes to pick the correct ones, kind of like subway surfers). Each round is 45 seconds, if you are doing well throughout your round, the game will speed up, if you are getting a lot wrong, it will slow down. If you don’t pick the correct items by the end and don’t have a high score, you receive 1 or 2 stars, which results in a failure. 3-5 stars means you passed, but you are able to beat your highscores in every round. I chose this idea because I like to play games that are fashion-related and ones that have a best score that you need to beat. It’s not too repetitive because it will shuffle through 6 different prompts (for now), and there are a lot of items to choose from that will work for each one. One of the design elements I like is that it will slow down if you are messing up a lot. In my opinion, if a game is moving too fast I get annoyed and close out, but the speed is adaptive to the player’s ability. Another thing I like is how the background changes based on where the prompt is. For example if you are getting ready for the gym, the obstacles will be dumbbells, and background will be a gym.

One design element that Claude created that I disliked was that it would put each item as red or green based on whether it was correct or not. This defeats the purpose of the game because it shouldn’t tell you which ones to pick.

Claude and I collaborated a lot throughout the process. I was able to communicate my idea in the way I was thinking about it in my mind, then have Chat create well-written prompts for the code. I went back and forth giving my corrections, but after I thought I was done, I asked Claude Code if there was anything else that it thought needed improvement. It made my character more detailed and added pause and mute buttons.

My strategy was to give it fewer prompts and corrections, but make them long and

detailed. I did run out of Claude credits on my Northeastern account after my first draft of the game, but I had previously requested a premium seat, so I was able to continue. Sonnet 5.5 worked the best for me to create my game, while Opus 5.5 was much slower.

AI Disclosure

I used Claude to build this game: Claude Chat for writing the detailed revision prompts and Claude Code (Opus 5.5, later Sonnet 5.5) to write and edit the code. Claude wrote all of the code. I did not write any myself. I made the design decisions and tested the game by playing it. There were about 12-15 rounds of revision in total. 1-2 to build the first version and about 11 in the second stage. Each round I played the game, described what was wrong, and Claude changed it.The first version looked too simple and cartoonish, so I asked for a full visual redesign of the character, clothing and backgrounds. Round time and prompt timing were wrong: 60 s was too long and the event prompt vanished too fast. I changed them to 45 s and 5.5 s with a countdown. Text was too small and the game moved too fast to read, so I asked for bigger text and a slower speed. Pop-up messages stayed on screen too long and covered the next items. Claude's first

speed system sped up just because time passed. I replaced it with speed that rises only when you're doing well. I asked for colour hints showing good and bad items, then removed them because they gave away the point of the game. Some rows had obstacles plus a wrong item and no good choice. I had Claude fix that. The first music loop was the same for every occasion, so I had it replaced with a different style for each event. Claude checked its own work by running the game in a browser and simulating rounds. I played it myself to decide whether it was fun and fair.