Gemini AI OCR
Google Gemini 1.5 Flash Vision extracts Sinhala text from scanned PDF pages with high accuracy.
UNICEF ADT Initiative · Sri Lanka
Making Sinhala Textbooks Accessible for Every Child
An AI-powered proof of concept that converts Sri Lanka Ministry of Education Sinhala PDF textbooks into WCAG 2.1 AA compliant digital textbooks — with TTS, keyboard navigation, and EPUB3 output.
WCAG 2.1 AA
EPUB3
Gemini AI
Open Source
Built for UNICEF's ADT Initiative
UNICEF's Accessible Digital Textbooks (ADT) initiative ensures children with disabilities have equal access to quality education in their native language. In Sri Lanka, over 80,000 children have visual or hearing impairments with severely limited access to accessible Sinhala educational materials.
This proof of concept demonstrates how AI — specifically Google's Gemini 1.5 Flash for OCR and MiniMax M2.5 for text structuring — can automate the conversion of existing Ministry of Education PDF textbooks into fully accessible WCAG 2.1 AA compliant digital formats at scale.
Learn about UNICEF ADTHow It Works
Three steps from any Ministry of Education Sinhala PDF to a fully accessible digital textbook
Upload PDF
Upload any Sri Lanka Ministry of Education Sinhala textbook — text-based or scanned. Supports grades 1–13.
AI Processing
Gemini 1.5 Flash OCR extracts Sinhala text from scanned pages. MiniMax M2.5 structures chapters and generates accessible summaries.
Accessible Output
Download WCAG 2.1 AA EPUB3 for screen readers, or read instantly in our accessible web reader with TTS and keyboard navigation.
Accessibility Features
Built to WCAG 2.1 Level AA — every feature serves a real accessibility need
Sinhala TTS
Web Speech API with si-LK voice — reads Sinhala text aloud. Gracefully falls back to default voice if si-LK unavailable.
Keyboard Navigation
Full keyboard control — Space to play/pause TTS, arrow keys for chapters, +/- for font size. No mouse required.
High Contrast Mode
Toggle high contrast for users with low vision. Black background with white text, meeting WCAG AAA contrast ratios.
Adjustable Typography
Four font sizes (S/M/L/XL) using Noto Sans Sinhala — the gold standard for Sinhala script rendering.
EPUB3 Output
Generates WCAG 2.1 AA compliant EPUB3 with accessibility metadata for screen readers like JAWS and NVDA.
Technology Stack
Python 3 + PyMuPDF
Gemini 1.5 Flash
MiniMax M2.5
ebooklib EPUB3
Web Speech API
Docker
Noto Sans Sinhala
WCAG 2.1 AA
Try the Web Reader
Built-in Sinhala demo content — works without a backend
පාඩම 1 — ශබ්ද හඳුනා ගනිමු
අපේ රටේ ලස්සන ස්වභාවය ගැන ඉගෙන ගනිමු...
Open Web Reader
Sample content included — no PDF upload required to explore the reader