Back To All Project
AniDraw is an Android educational drawing game designed to teach children five animal types (cat, cow, fish, dolphin, duck) through interactive digital drawing, powered by an on-device AI sketch classification system. Developed as a Capstone Project for the D4 Software Engineering Technology program at Politeknik Negeri Padang.
Built as a full end-to-end system spanning three components: a Flutter mobile app (Clean Architecture with BLoC/Cubit), a CNN-based sketch classifier (MobileNetV2 via transfer learning, quantized to TFLite INT8 for offline on-device inference), and a Node.js/Express backend with PostgreSQL for data, authentication, and asset management, plus a Next.js admin panel for content and analytics management.
The classification model was trained on 15,000 sketch images (Quick, Draw!-derived dataset) and evaluated on an independent test set, achieving 88.53% accuracy. The system passed full blackbox testing (13/13 scenarios) and was validated by kindergarten teachers and early-childhood education students, who rated it as feasible for use with minor revisions.
Solo-built a full 3-tier system spanning Flutter mobile app
ML pipeline
and backend/admin services
Trained a CNN (MobileNetV2) sketch classifier achieving 88.53% test accuracy
deployed on-device via TFLite INT8 quantization
Designed gamified learning features including scoring
shop
inventory
and leaderboard systems
Validated through 100% blackbox test pass rate and expert UAT by kindergarten teachers and PGSD students
RoleSolo Developer (Mobile, Machine Learning, Backend, and Admin Panel)
Statuscompleted
Tech Stack
Flutter
Dart
BLoC
Python
TensorFlow/Keras
TensorFlow Lite
Node.js
Express.js
PostgreSQL
Prisma
Next.js
React
Cloudflare R2