Aksara Ulu is an ancient script used across South Sumatera for centuries — inscribed on bamboo, bone, and wood. Today, very few people can read it. AKAS was our attempt to change that.
The Problem
Aksara Ulu has no standardized digital keyboard. Most young people in South Sumatera have never seen it. Academic resources exist in physical archives, largely inaccessible. The script risks becoming purely a museum artifact.
Our Technical Approach
We built two core features: a transliteration engine (Aksara Ulu ↔ Latin) and an image recognition model that identifies handwritten characters from photos.
- Python + Keras CNN model trained on 2,400 character samples across 56 character classes
- Laravel API serving both transliteration rules and image recognition inference
- Kotlin Android app with CameraX for real-time character capture
- Offline-first transliteration — works without internet connection
The Hardest Technical Challenge
Training the image recognition model was the toughest part. Aksara Ulu has significant regional variations — the same character is written differently across districts. We had to collect samples from multiple regions and build a dataset from scratch with local historians.
# Simplified model architecture
model = Sequential([
Conv2D(32, (3,3), activation='relu', input_shape=(64,64,1)),
MaxPooling2D(2,2),
Conv2D(64, (3,3), activation='relu'),
MaxPooling2D(2,2),
Flatten(),
Dense(128, activation='relu'),
Dropout(0.3),
Dense(56, activation='softmax') # 56 character classes
])
💡 We achieved 87% recognition accuracy on the test set. Not perfect — but good enough for an educational tool where users can see multiple suggestions and confirm the right one.
Impact
The app has been downloaded by schools across South Sumatera and featured by the provincial culture department. More importantly, teenagers are using it to decode old family documents — which is exactly what we hoped for.