How to build your own Rust app with AI coding
My wife looked at the chaos on her hard drive and asked if I couldn't just build an app to solve it. Multiple import folders filled with thousands of photos, videos, and endless duplicates. Sorting manually would take her days.
Now, I'm absolutely not a programmer, but I've been experimenting with vibe coding for a few weeks. This seemed like the perfect project to see how far that would take me.
Why I chose Rust
I could have accomplished this with Python, but I'd heard many good things about Rust. That language is known for speed and efficient memory usage, but especially for preventing errors that often get in the way of AI driven development. And I'd long wanted to create a real Windows executable, so this was the moment.
Rust normally has a steep learning curve for developers. But the combination of Antigravity and Gemini worked surprisingly well. I could build a stable tool quickly even without a technical background.
The design process upfront
Before I started writing code, I invested time in the design. I used Gemini as an assistant for this. We worked through what the tool needed to do exactly, what the interface looked like, which functions were essential. That groundwork proved really worthwhile later.
The first version did exactly what I had in mind. Moving files to folders by year, setting duplicates aside separately, everything worked immediately as intended. Really remarkable.
Keeping focus on what was truly needed
I considered adding a gallery view within the app, a visual overview of all organized photos. But after a few attempts, I decided not to pursue that. The chosen GUI didn't seem like the right choice for that purpose. More importantly, it wasn't really necessary. Viewing and further cleaning up the photos works excellently from Windows Explorer with preview view.
Exactly what building your own tool is about. You don't need to add features because they "should be there." You only build what solves the problem.
From personal tool to public release
Because the main goal worked so well, I decided to publish the tool on GitHub. That decision led to a number of additional iterations. Adding a warning message that it's wise to have a backup and that use is at your own risk. Clear names for the buttons. A proper icon. And extra focus on user-friendliness for people opening the tool for the first time.
The iterations were all small improvements. I never had to solve errors or problems. I think that's an advantage of Rust in combination with AI tools, it prevents many of the typical beginner mistakes that normally occur.
Compiling in WSL to Windows executable
What I also found remarkable is that I did the compiling in WSL (Windows Subsystem for Linux) and the output is a Windows application. Initially I also got a Linux version, but we abandoned that.
Assigning the icon to the executable took some time. That didn't work well from WSL, so we saved the icon we use within the app separately. I could manually convert this to the correct icon format. After each compilation, I have to assign the icon to the exe with a tool within Windows via PowerShell. A bit of manual work, but that's no problem for me since I expect few changes.
The 10,000-file test
We used my own photo archive as the first big test. In just a few minutes, five different import folders with nearly 10,000 files were completely organized. What would normally be an endless chore in Windows Explorer was done with a few clicks. The duplicates were neatly collected and only needed a final check.
The bigger picture
The entire project took about 24 hours, including all the steps to publish it as a release on GitHub. For me, this was a wonderful project to discover how powerful vibe coding is for solving practical problems, even if you can't write a single line of code yourself.
But more importantly, this illustrates a new form of digital autonomy. When you identify a manageable digital problem, you now have a choice. You can search for an existing tool, but you don't know upfront whether it does exactly what you want. Creating folders by year, finding duplicates, you'd have to test all that. And then you often get all kinds of bells and whistles that you don't need at all.
Or you build a tool yourself that only focuses on your specific problem. No compromises, no unnecessary complexity. Only what's needed to solve the problem.
That's perhaps the most beautiful form of digital autonomy for me. You have control over your data, and over the tools you use to manage that data.
Do you have this photo chaos too?
If you recognize this chaos and keep putting off the manual organizing, you can find the tool here. The complete source code is also on GitHub, so you can see exactly what the tool does before you use it.
And have you identified a small digital problem yourself for which you might be able to build a simple tool? I'd love to hear about it. Let's share ideas on Reddit.
Download here the Photo & Video Organizer
Share your ideas in our Reddit community
About the author
Peter van Barneveld is a Group Innovation Manager exploring practical paths to digital autonomy. He tests what actually works in real daily use. His approach: figure it out, document the journey, share honest results.
Connect with Peter on LinkedIn or join the discussion on r/Dimaginar.
Frequently asked questions
Do you really need to know nothing about programming?
What you do need is patience to go through the design process with an AI assistant, and willingness to learn to understand what the tool does. You need some understanding of technical concepts, but no coding expertise. The AI writes the code, you steer the decisions and evaluate the results.
What do you concretely need for such a project?
An AI coding agent, a development environment like WSL with VS Code, and an AI assistant for questions. Time to learn and experiment. Always critically evaluate what the AI suggests, especially with code and system changes.
What does this actually cost in AI tools?
For this project I only invested time, no money. I used Antigravity, which is generous with free credits in the preview phase. Gemini 3 Pro for the coding plan and initial output, then Gemini 3 Flash for iterations. For larger changes sometimes Claude Sonnet 4.5 (also via Antigravity). For help I used the free Gemini version. And GitHub for publication is also free.
What if I find a bug or have a suggestion?
Open an issue on GitHub, then we can address it via GitHub features. Or come to the Reddit community to share ideas. This is my first public tool, so feedback is welcome.