Visually and hearing impaired people in India know a banknote's value in under 1.5 seconds, by voice or vibration, with no internet.
RBI wanted people who cannot see or hear well to identify every banknote on their own, because the raised marks on notes wear away with use.
What changed
What stood in the way.
- Tactile marks on notes from INR 10 to 2,000 wear away, so people relying on touch could not tell denominations apart.
- The app had to work offline, on worn and folded notes in poor light, in many languages, and stay small and secure.
- 1M+downloads (as reported)
- 99.9%accuracy (as reported)
- Under 1.5 sto recognise a note (as reported)
What we built.
- A dataset of over 150,000 images of Mahatma Gandhi and Mahatma Gandhi (New) Series notes, in different orientations, light, folds, cameras and backgrounds.
- TensorFlow models trained by transfer learning from ImageNet, with Tesseract OCR in training, converted to run on the phone without internet, with automatic flashlight.
- Android (Java) and iOS (Swift) apps that announce the value in 11 languages, or by a set number of vibrations for each denomination.
- Voice navigation through Siri and TalkBack, support by SMS or missed call with a support dashboard, and VAPT and SAST security testing.
- TensorFlow
- Transfer learning (ImageNet)
- Tesseract OCR
- Java (Android)
- Swift (iOS)
- SQLite
What this work shows about us.
Voice in their language, vibration for those who cannot hear, and no internet needed, because that is how the app is used.
Training covered worn, folded and badly lit notes, the cases where touch alone fails.
Want to build a similar banknote recognition app? Tell us what you want to change. Our AI assistant, Ask Unthinkable, helps you define your requirements, then connects you with the right team.
Start your plan ↗If we built it today, with AI.
What we would add now, moment by moment.
An agent would answer support requests sent by SMS or missed call in the caller's language, and pass unresolved ones to the support team. People handle every complaint.
An agent would group feedback and failed scans by note and condition, so the team knows which cases to add to training. The team decides each update.
Where this work fits.
Facts and figures from the project records; nothing added.
Want this for your business?
Tell us what you run or what you're building. Our AI assistant, Ask Unthinkable, helps you define your requirements, then connects you with a senior AI engineer.
Send us a note instead
An AI agent. This prototype uses prepared answers and sends nothing.
We use essential cookies to run this site. With your consent we also use analytics cookies to see which pages help. Privacy policy