This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady
What I Built
Digital Janitor is a local‑first, AI‑powered assistant that removes digital clutter from your machine.
What it does – Users point the app at any folder (or the whole home directory). The backend recursively scans every file, detects exact duplicates, flags ambiguous filenames, and extracts lightweight metadata (size, type, timestamps). An on‑device LLM (run via Ollama) then generates semantic, human‑readable filenames and suggests a clean folder hierarchy. All proposals are shown in an interactive dashboard where you can accept, tweak, or reject individual changes before applying them.
Effort saved – Eliminates hours of manual file hunting, renaming, and duplicate removal.
Who gets lazier – Students juggling research PDFs, designers with sprawling asset libraries, remote workers maintaining shared drives, and anyone who simply wants a tidy computer without spending a weekend reorganizing.
Code
Github URL - https://github.com/Praveen-Raj-Thulasi/DigitalJanitor
How I Built It
Open‑weight AI model – I use the Llama 3.2 model (7B parameters) via Ollama, running completely offline on the user’s machine. This keeps data private and eliminates any usage‑based costs.
Frameworks –
Backend: Express, Mongoose (MongoDB for persistent metadata), and express-rate-limit for basic protection.
Frontend: Vite + React + Tailwind‑in‑CSS (custom CSS for premium aesthetics).
AI integration: A thin wrapper service (/api/ai) forwards file‑content snippets to Ollama, receives suggested filenames and hierarchy JSON, and returns it to the UI.
Local‑first architecture – The scanning and AI inference happen on the client‑side (via the browser’s File System Access API) or on a tiny local Node process, so no data ever leaves the device.
Testing – Vitest + Supertest cover the API endpoints (auth, scans, cleanup plans) and the filesystem utilities, ensuring the app works end‑to‑end in the test environment (NODE_ENV=test).
Why Does Open Innovation Matter?
Open‑source AI models let anyone run powerful language‑understanding locally without paying per‑request fees or trusting a proprietary service with personal files. By building Digital Janitor on Llama 3.2 and exposing the inference pipeline as an open API, the project:
Democratizes access – Users on modest hardware can still benefit from intelligent filename suggestions.
Enables community extensions – Anyone can swap the model for a domain‑specific one (e.g., scientific paper summarizer) or add custom rules for duplicate detection.
Protects privacy – No cloud endpoint means no hidden data collection, a crucial advantage for sensitive documents.
Accelerates adoption – Because the code is MIT‑licensed, developers can fork, contribute, and integrate the cleanup engine into other tools (IDE plugins, cloud sync services, etc.) without legal friction.
Open innovation turned a closed‑API‑costly idea (“Ask an external service to rename my files”) into a free, privacy‑first solution that anyone can run, improve, and deploy.
Top comments (0)