Confere: Local Receipt Reading With a Paper Trail

Описание и применение

Прототип Confere позволяет фрилансерам в Бразилии обрабатывать фотографии чеков локально, проверять OCR‑вывод, устранять дубликаты и экспортировать подтверждающие пакеты с исходными изображениями и хешами.

Задача

Freelancers need a trustworthy, inspectable way to convert receipt photos into approved expenses without re‑entry or hidden inference costs.

Результат

The prototype processes three sample receipts locally in under a second, correctly identifies totals, blocks duplicate approvals, and produces an evidence packet that can be independently verified.

Рабочий процесс

• Add PNG, JPG or WebP images
• Inspect source image, correct merchant/date/BRL amount and choose a category
• Resolve possible duplicates before approval
• Export a CSV or a self‑contained evidence packet containing the original images, recognition text, corrections, source‑image hashes and review timestamps

Доказательства

In real browser inference on clean synthetic images, the café's source total 34,90 and stationery shop's 124,90 were both read and parsed correctly.

Источник

The duplicate failed approval until resolved. Two approved originals produced exactly 159,80.

Источник

Eleven unit tests cover the calculation and validation boundaries.

Источник