Raw ·
Receipt parser that runs entirely on your phone
Snap a photo of a receipt; an on-device model extracts merchant, amount, category, and date. Nothing leaves the phone.
The Problem
Every expense app sends your receipts to someone else's server. Most users would prefer to keep that data local but the available apps don't offer the option.
The Solution
A small vision-language model fine-tuned on receipts, packaged as an iOS/Android app. Runs offline. Exports to a CSV or syncs to a self-hosted endpoint of your choice.
The Opportunity
Privacy-first finance is an underserved niche. The same model can extend to invoices, business cards, ID documents — anything you'd rather not upload.
Technical notes
- On-device VLMs are now small enough (~1B params quantized) to run on a flagship phone in <2s per image.
- Categorization should be user-editable; learned per-user, not server-side.