Inspiration

Every day, we waste hours organizing files. Downloads pile up. Screenshots scatter. Important documents get lost. I built AutoDrop to solve my own digital chaos.

The idea: What if AI could understand what's IN a file—not just its extension—and instantly organize it to the right place?

What it does

AutoDrop is a Windows desktop app with a floating drop zone. Drag files onto it, and Google Gemini AI analyzes the content:

  • Images: Gemini Vision identifies screenshots, receipts, photos, diagrams
  • Documents: Gemini reads PDFs and text files to understand context
  • Smart Suggestions: AI recommends the perfect folder based on content
  • One-Click Organization: Select a destination, file moves instantly
  • Full Undo History: Every operation is reversible

How I built it

Architecture:

  • WPF (.NET 8) with Fluent Design UI
  • MVVM pattern with dependency injection
  • Multi-provider AI architecture (Strategy Pattern)

Gemini Integration:

  • Gemini 3 as primary AI
  • Vision API for image content analysis
  • Text API with structured JSON output for categorization
  • Prompt engineering for consistent, actionable responses

Key Technical Decisions:

  • SOLID principles throughout
  • Interface-based services for testability
  • Secure API key storage with Windows DPAPI
  • Local AI fallback for privacy-first users

Challenges I ran into

  1. Multimodal Prompts: Getting Gemini to return consistent JSON for both images and documents required careful prompt engineering

  2. Rate Limiting: Implemented request queuing for batch file operations

  3. File Type Detection: Some files (screenshots vs photos) needed content analysis, not just extension checks

  4. Balancing Speed & Accuracy: Used Gemini Flash for speed while maintaining high accuracy

Accomplishments that I'm proud of

  • Gemini 3 Integration: Successfully using the latest Gemini models to "see" and "read" files locally on the desktop.
  • Professional Polish: The app has a full installer (Inno Setup) and a settings manager that rivals commercial software.
  • Privacy First: I implemented a "Local AI" fallback using ONNX models for users who want offline organization, alongside the cloud-powered Gemini integration.

What I learned

  • Gemini's vision capabilities are exceptional for real-world image categorization
  • Structured JSON responses from LLMs require explicit prompting
  • The importance of fallback strategies (local AI when cloud unavailable)
  • How to architect a system for multiple AI providers

What's next

  • Folder monitoring for automatic organization
  • Cloud sync support
  • Microsoft Store distribution

Built With

  • .net-8
  • c#
  • communitytoolkit.mvvm
  • dpapi
  • gemini-2.5-flash
  • gemini-3.0
  • gemini-3.0-flash
  • github-actions
  • google-gemini-api
  • inno-setup
  • moq
  • mvvm
  • serilog
  • windows
  • wpf
  • wpf-ui-(fluent-design)
  • xunit
Share this project:

Updates