Jump to an app, a write-up or an action
The goal was a genuinely useful receipt scanner that costs nothing to run, so the heavy lifting happens on the visitor's device.


Tesseract.js runs in a Web Worker and is only downloaded when someone scans. Before reading, the photo is downscaled and re-encoded on a canvas, which keeps uploads under the 1 MB Server Action limit and strips EXIF data such as GPS location.
A parser turns noisy OCR text into a merchant, total, date and category. It prefers "grand total" over "total", ignores subtotals, tax and cash tendered, and handles US and day-first dates. You always review the result before saving; "What was read" shows the raw text.
View the codeReceipts are stored with private access and never linked directly. A route streams each image only after checking that the signed-in user owns the expense. Uploads are identified by their magic bytes rather than the type the browser claims.
Money is stored as integer cents. The CSV export prefixes values that start with =, +, - or @ so a malicious merchant name can't run as a spreadsheet formula.
The charts are plain HTML and CSS, built to a written data-viz spec. Colors were checked with a contrast and color-blindness validator against the actual card backgrounds in both themes, and every chart has hover and keyboard tooltips plus a data table view.
View the code