Back to the world

AI-Assisted Desktop Tooling

2025 · Shipped

SmartFileOrganizer

SmartFileOrganizer is a Python desktop application that brings order to chaotic directories. It combines deterministic rule-based sorting with AI-assisted classification, wrapped in a native-feeling PySide6 interface.

GitHub ↗

Problem

Downloads folders become landfills. Operating system tooling offers only manual dragging and primitive name sorts, while people think about files semantically — invoices, notes, screenshots, projects.

Solution

A watch-and-organize desktop app. Deterministic rules handle known patterns instantly; ambiguous files are classified by content-aware heuristics with user confirmation. Every action is logged and reversible in one click.

  • Live folder watching with debounced organization passes
  • Rule engine with extension, pattern, size and date conditions
  • AI-assisted categorization for ambiguous files, always confirm-first
  • Full activity log with undo for every move operation
  • Profiles for different workflows (downloads, documents, media)

System architecture

  • frontend

    PySide6 UI

    Main window, tray icon, settings and confirmation dialogs.

  • backend

    Folder Watcher

    Filesystem observer feeding a debounce queue.

  • backend

    Organization Engine

    Rules pipeline producing planned file operations.

  • ai

    Classifier

    Content heuristics plus AI calls for low-confidence cases.

  • data

    Local Store

    SQLite journal of every operation enabling undo.

Data flow

  • Folder WatcherOrganization Engine
  • Organization EngineClassifier(low confidence)
  • ClassifierPySide6 UI(confirm)
  • Organization EngineLocal Store
  • PySide6 UIOrganization Engine

Technology

  • Python
  • PySide6
  • AI APIs
  • Automation

Engineering challenges

  • Bulk moves triggered re-scanning storms as organized files re-entered watched folders.

    Added operation identity tagging so engine-produced moves are ignored by the watcher, plus hysteresis windows on event bursts.

  • Wrong AI classifications eroded trust quickly.

    Redesigned the flow around suggestions: nothing moves without approval unless the rule confidence exceeds an explicit user threshold.

Decisions & trade-offs

  • Confirm-first autonomy dial

    Trust in automation is earned incrementally; users should graduate from approvals to hands-off.

  • SQLite operation journal

    Undo is a first-class feature for anything that touches a user's files.

  • Deterministic rules before AI

    Predictable behavior for the 90% case reserves AI cost for genuine ambiguity.

Result

  • A dependable desktop utility used daily on my own machine.
  • Zero-data-loss record across all organization runs thanks to the journal-and-undo design.
  • A clean example of blending classic automation with modern AI classification.

What I learned

  • Desktop software lives or dies by its undo button.
  • Watch services must assume their own actions will re-enter the pipeline.
  • AI features need UX for disagreement, not just accuracy.