Doze Sort uses a local model to file notes into copied folders, keeping originals

Doze Sort is a small open-source tool built for a Hacktoberfest AI challenge that uses an open-weight model to sort a pile of text or Markdown notes by comparing their contents to topic descriptions. Users review a plan before one batch action creates organized copies in proposed folders, while uncertain or low-scoring notes go to a review folder. Original files are never removed, and byte-identical duplicates point to a single canonical copy. The classifier uses the all-MiniLM-L6-v2 model, with heuristic thresholds that can still miss mixed-topic notes, and undo moves a completed run into recovery without deleting files. The author notes the system assumes a stable local filesystem and is not designed for hostile concurrent writes.