Files
grocery-management-obsidian/Receipt Scanning.md

1.1 KiB

OCR

  1. User takes a picture of the receipt
  2. Receipt image is uploaded
    1. Document AI can do direct (online) or batch ("offline") processing
    2. Need to see how we want to handle it, but it could upload to API and have API talk to Document AI, or have app talk to Document AI. Leaning toward the former.
  3. Image is processed, results stored in datastore
    1. Probably should attempt to scrub payment info from the data

Line item categorization

Have had several thoughts on this. Initial thought was to have a category/inventory item with a list of assigned aliases, but this feels like it could become unwieldy. Next thought was to maintain a list of inventory items, then when processing the receipt run a fuzzy search against the list to see what comes back. A list could (and probably should) still be maintained, but it would be a fallback if fuzzy search returns nothing (albeit more expensive, since it'd have to check every list) Maybe relax the fuzzy search to narrow which fields to scan the lists of?

Unaddressed items

  • How to handle voided items?