-
-
It's First Prototype
Inspiration
The inspiration came from the Forest app where you can grow virtual trees based on how long you focused. I wanted something similar that would show my progress for my goals and various aspects of my life. I didn't really know what UI to go with but I wondered if GPT 5.6 Sol would be able to render 3D islands in React Native/Expo and it did it successfully. So that's how the UI and design was born.
What it does
You first sign up/in and begin an onboarding chat with GPT 5.6 Luna who tries to understand what aspect of life you're focusing on. It tries to understand your goals or aims for topics such as Fitness, Sleep, Work, etc.. and it creates islands for each topic. Then you can check in everyday and discuss your progress with Luna in the chat or complete the personalized questionnaire for each island. If you do well, your island automatically grows and if you do something negative against your goals, you could slowly accumulate rocks that sink your island.
How we built it
I built it using Codex and GPT 5.6 Sol High as the main orchestrator and GPT 5.6 Sol Low as subagents. I initially had a brainstorming session, then had the model prototype the 3D map and then set up the documentation and git workflow so the model could be as independent as possible. To be honest, I was very busy during this period so I didn't have time to test and even try the app very often (in fact I don't even have an iphone ironically). So I gave it a lot of free reign and even opened my convex, clerk and apple developer accounts so it could set everything up and test everything end to end. While it took a long time, forgot one feature, and burnt a lot of tokens, it still returned successfully.
Challenges we ran into
Debugging animations was tricky because I didn't want to be in a constant back and forth loop with GPT. Instead, I told it to record a video of itself using the app and debug the animations from the videos. It mostly worked, except the animations were always much slower than expected. But that was an easy fix compared to debugging weird glitches in animations by hand.
Accomplishments that we're proud of
I'm proud of GPT for carrying me and finishing the app. (Eventhough it cost me all 4 of my usage resets). And I'm proud of myself for finishing this hackathon.
What we learned
GPT 5.6 is much more independent but it is a token burner. When using the Chrome and Computer Use plugins it can do pretty much anything. However, the token costs need to be optimized further. I did set it to only use GPT 5.6 Sol Low reasoning for subagents but it could be worth using even smaller models like Luna and Terra for this.
What's next for Mindland
I'd like to make the onboarding flow smoother and I'll aim to deploy it on App Store.
Log in or sign up for Devpost to join the conversation.