Submit to the week 3 DEV Challenge(hactoberfest)

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

LazyDesk is a local-first, privacy-respecting desktop application that scans messy folders (like your overflowing Downloads or cluttered Desktop), computes an intelligent category-based structure, and moves only the files you explicitly approve.

Most AI file tools expect you to hand over access to your private folders and upload documents to third-party cloud APIs. LazyDesk takes the "Slow and Steady" approach:

Your files never leave your disk: File contents are never uploaded to remote servers. When cloud AI is enabled, only file names and extensions are evaluated.
Review-First & Zero Overwrites: AI proposes moves; it never executes them in the dark. Every conflict is auto-renamed or skipped, never overwritten.
Instant 1-Click Undo: Every operation is logged in an ACID SQLite transaction. If you change your mind, hit Undo and your files return to their exact original locations safely.
Deterministic Fallback: For obvious files (.jpg, .mp4, .pdf), deterministic rule-matching works with 0 tokens and 0 latency. Open-weight AI is reserved for ambiguous filenames like tax_return_final_v2 or project_draft_october.
Zero Cost & Free Tiers: Fully functional with local open-weight models via Ollama (qwen2.5, llama3.2, gemma2), free hosted inference (Pollinations, Groq, OpenRouter, Google AI Studio), or in 100% offline rules-only mode.
Whoever has an unmanageable Downloads directory can finally sit back, be a little lazier, and let LazyDesk do the boring chores.

Demo

Live Hosted Demo: https://lazydesk-web.onrender.com

Code

github link : https://github.com/Nareshkumawat-star/lazy-desk

How I Built It

LazyDesk is engineered as a secure two-tier local system wrapped in a native desktop shell:
Electron Desktop Shell (Context Isolation, Sandbox, Native Picker)
│
├──► Next.js 15 UI (React, Tailwind CSS, Lucide Icons, TypeScript)
│ │ (Typed HTTP + Loopback + X-LazyDesk-Token Header)
│ ▼
└──► FastAPI Local Service (Python 3.12, 127.0.0.1:8765)
├── Categorization Engine (Rules + Open-Weight Model Registry)
├── Collision-Safe Operation Planner (shutil.move with validation)
├── SHA-256 Deduplication Scanner
└── SQLite Storage (Sessions, Operations, Embeddings, Rules)

1. Open-Source AI Stack & Local Inference Local Inference via Ollama: Supports private on-device models such as qwen2.5:7b, llama3.2:3b, phi3:mini, and gemma2:2b. Semantic search embeddings use nomic-embed-text. Zero-Setup Free Cloud Inference: Defaults to Pollinations (gpt-oss-20b) without any API key or signup required, with optional adapters for Groq (Llama/Qwen), OpenRouter free open-weight tiers, and Hugging Face. Structured JSON Extraction with Pydantic: Model responses are treated as untrusted text. Outputs are parsed and validated strictly against defined categories to prevent prompt hallucinations or path injection.
2. Defensive Filesystem Safety Symlinks are never traversed during scans or moves. Directory traversals (../) are actively blocked by strict boundary validators. Pre-move and pre-undo validation passes verify that source files still exist and target destinations are completely collision-safe before touching disk. ## Why Does Open Innovation Matter? When dealing with a user's personal desktop, privacy is paramount. A traditional proprietary tool demands subscriptions, ties you to proprietary cloud endpoints, and uploads your documents to black-box commercial APIs.

Open innovation made LazyDesk possible in ways closed APIs cannot:

True Local Privacy: Open-weight models (Qwen, Llama, Gemma) allow complex natural-language file interpretation to happen entirely inside the user's laptop RAM without sending a single byte outside.
No Paywalls or Vendor Lock-in: By decoupling the organization logic from proprietary APIs, anyone can run LazyDesk forever with zero operational costs.
Verifiable Auditability: Because the models, FastAPI backend, and Electron preload bridges are open-source, users and security researchers can verify that their personal files remain strictly on their hard drive.

Prize Categories

Best Use of Render: LazyDesk includes a production-ready, schema-validated render.yml Render Blueprint that provisions both the Python FastAPI service and the exported Next.js static front-end with automatic service token wiring, explicit CORS origin restrictions, and health check probes.
Best Use of Gemma: Supports local and hosted Gemma 2 (gemma2:2b) inference for efficient on-device categorization and lightweight CPU/GPU setups.

Team leader (username):nareshkumawatstar