This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady
What I Built
Meet SlowSort 🦥, a local-first file organizer that helps you spend less time cleaning up messy folders.
We all have files named IMG_20261012.png, invoice_final_v3.pdf, or download_001.jpg. Sorting them manually means creating folders, renaming files, and repeating the same work again and again.
SlowSort makes that workflow easier. It scans file metadata, suggests categories and cleaner filenames, and prepares an organization plan for you to review before anything changes.
What it can do
- Suggest categories and destination folders for files.
- Recommend clearer filenames.
- Preview proposed changes before applying them.
- Let users review and approve individual operations.
- Handle filename conflicts safely.
- Maintain an activity log and support undo for eligible operations.
- Run a deterministic demo mode without requiring an AI model.
The idea is simple: let AI handle the boring sorting while you stay in control.
Demo
Explore the project through the screenshots below and the source code on GitHub.
The repository includes setup instructions, screenshots, automated tests, and details of the local-first file organization workflow.
🔗 GitHub: https://github.com/NirajDN/SlowSort
Code
🔗 GitHub repository: https://github.com/NirajDN/SlowSort
The repository contains the React frontend, Express backend, file-safety services, automated tests, setup instructions, and screenshots.
Screenshots
1. Dashboard
2. Suggestions and review
3. Completed organization
How I Built It
I built SlowSort with React 18, Vite, Node.js, Express, and Lucide icons.
The backend separates scanning, classification, suggestion generation, path validation, file operations, and activity logging.
The project includes an Ollama integration configured for a small open-weight model, gemma3:1b, alongside a deterministic rule-based fallback. The fallback lets users explore the workflow without installing or running an AI model.
Safety was a major design consideration. File operations are restricted to the demo sandbox, suggestions are reviewed before execution, and filename conflicts are handled without blindly overwriting existing files. The scanner uses metadata such as filenames, extensions, sizes, and modification times rather than reading document contents.
The reported automated test run passed all 25 tests, covering safety validation, classification, preview, apply, undo, and API behavior. The frontend production build also completed successfully.
Transparency note: The Gemma inference path has not yet been verified in a live run. The deterministic demo workflow and reported automated tests are separate from that unverified model integration.
Why Does Open Innovation Matter?
File organization can involve personal documents, work files, and other information that people may not want to send to an external service.
Open-weight models give developers more control over model selection, deployment, and experimentation. With local inference, an application can be designed to process the information it needs on the user's machine instead of requiring a hosted AI API.
Open innovation also makes the underlying approach easier to inspect, adapt, and improve. Developers can experiment with different models and build tools around their own needs.
For SlowSort, AI is responsible for suggesting an organization plan, while explicit application logic validates file operations and requires user approval. That separation is important: a model should not have unrestricted authority over someone's files.
The Lazy Payoff
Instead of manually planning every folder and filename change, a user can scan, review suggestions, preview the plan, and approve the desired operations.
The automated tests verify important parts of the demo workflow, including sandboxed file operations and supported undo behavior. I have not conducted a timed user study, so I am not claiming a specific number of minutes saved.
The goal is to make repetitive digital housekeeping less tedious without sacrificing privacy or control.
Prize Categories
No partner-specific prize category claimed at this time.
Top comments (0)