FacetDrive

The idea for FacetDrive originated during a conversation in ChatGPT Work about the OpenAI Build Week. My initial concept was to make the ChatGPT library accessible as a virtual drive on Windows. After realizing that it would conflict with the ToS, I discarded it and started exploring a different direction. I thought about how disorganized the files on my own computer often become and came up with the idea of an application that creates a virtual drive and uses AI to organize them automatically. For the classification task, I chose GPT-5.6 Luna because it is a small, fast model that is well suited to this kind of workload. The implementation plan for the project was developed by me together with GPT-5.6 Sol in ChatGPT Work. We defined the main concept, architecture, requirements, technical approach and many of the key implementation decisions. I then reviewed and refined that plan before giving it to Codex, which used it as the foundation for building the project. One aspect that surprised me about GPT-5.6 Sol was how much easier it became to implement new features, ask for advice and request code reviews. In many cases, I would describe a feature from my phone while I was away from my laptop. By the time I could test it on my computer, the implementation was already very close to what I wanted, often requiring only a few small adjustments. Accessibility was one of my main priorities. I'm visually impaired and use a screen reader, so I paid particular attention to elements such as lists, checkboxes, dialogs, keyboard navigation, focus management, and accessible labels. I had previously developed and used other WPF applications, but I had often struggled to understand why certain controls were not properly exposed to assistive technologies. GPT-5.6 Sol was able to solve these accessibility problems successfully. I'm especially happy about this because the techniques and patterns developed for FacetDrive will also be useful in future projects built with the same framework. Because of my visual impairment, I also relied on GPT-5.6 Sol for assistance with the visual design. I still made an effort to understand and evaluate the interface myself, while using ChatGPT to describe images and other visual elements. I'd be grateful if the judges could take this context into consideration when evaluating the UI. Another part of the project that I particularly enjoyed was creating the video submission with Codex. I didn't expect it to be capable of producing the animations it created, especially the animated logo sequence, which became one of my favorite parts of the final result. For the voiceover, I experimented with a use case I had never seen before: using GPT Live 1 as a speaker, as though GPT Live 1 were behind the microphone and I were directing the performance, asking it to adjust its pacing, tone, emphasis, and delivery line by line. I wasn't sure this approach would work, but I think the final recording turned out pretty well.

Built With

• .net
• c#
• wpf