Papertrail Receipts

Inspiration

Paper receipts are easy to lose, difficult to search and awkward to preserve for returns or warranties. Many scanning services solve this by requiring an account or uploading sensitive purchase information. Papertrail was built to provide a useful receipt archive while keeping normal processing and durable storage on the user's Android device.

What it does

Papertrail captures receipts with the camera or imports them from the Android photo picker. On-device ML Kit OCR and a deterministic evidence engine suggest merchant, date, time, total, currency, payment method, organization number and receipt reference.

Instead of hiding uncertainty, Papertrail explains which fields need review and why. Users can correct fields, organize receipts into nested folders, search their archive, track warranties, create QR receipt copies, export selected records and create passphrase-encrypted backups.

No Papertrail account is required. The app contains no ads, cloud sync or runtime AI calls.

How we built it

The Android app uses React Native, Expo, TypeScript, Kotlin, ML Kit, SQLite and SQLCipher. Receipt interpretation combines OCR text, native line geometry, deterministic parsing, field-specific evidence policies, candidate adjudication and a shared review system.

How Codex and GPT-5.6 were used

Papertrail existed before Build Week. During the submission period, Codex with GPT-5.6 Sol helped deepen OCR geometry use, separate field-specific evidence policies, strengthen deterministic safety checks, create an end-to-end regression harness, improve confidence provenance, repair encrypted backup workflows, optimize receipt processing and prepare the Android judge release.

Codex accelerated implementation, investigation, testing and release verification. Human product decisions retained authority over privacy boundaries, extraction behavior, UX and which rules were sufficiently general to ship.

What we learned

Receipt extraction is not just OCR. Reliable results require provenance, geometry, competing candidates, suppression rules and honest review states. A transparent deterministic system can be more useful than a confident opaque answer.

What's next

Future work includes broader receipt-format evaluation, additional languages, improved QR checkout receipts, archive scaling and an eventual iOS release.

Built With

• android
• codex
• expo.io
• gpt-5.6
• jest
• kotlin
• ml-kit
• node.js
• python
• react-native
• sqlcipher
• sqlite
• typescript