Loading portfolio
Independent Mobile ProductIndependent Product · 3-person team

Duely

Founded and led a three-person team building an Android-first student organizer that turns assignment photos into editable tasks through on-device OCR and optional AI assistance.

GitHub
Duely project

/CASE STUDY

The challenge

Make it easier for students to turn handouts, screenshots, and photos into organized tasks without hiding OCR mistakes or sending every assignment image to the cloud.

Technical approach

  • Built the Android app with React Native, Expo Router, and TypeScript, including offline tasks, reminders, smart prioritization, calendar views, and local source-image retention.
  • Designed an on-device Google ML Kit OCR flow that shows uncertainty per field and requires an editable review before a task is saved.
  • Added optional Gemini-assisted extraction through an authenticated Supabase Edge Function with explicit consent, monthly limits, cancellation recovery, timeouts, and local OCR fallback.
  • Implemented Google sign-in, Row Level Security, cloud backup and restore, accessibility improvements, and a RevenueCat Test Store flow for simulated Duely Plus purchases.

Results

320 automated tests passing across 62 test files
24-case English, Filipino, and mixed-language parser evaluation
99.5% token accuracy across a controlled 12-image synthetic OCR pack
12/12 controlled scans produced parser-ready text
6/6 clear deadlines extracted, 3/3 missing deadlines avoided, and 3/3 ambiguous deadlines flagged
Android demo verified for OCR, optional AI review, task saving, and simulated Plus purchase and restore

/MORE PROJECTS