Center for Human-Machine Interaction Studies
|Research Assistant
Malang, Indonesia, Indonesia
Summary
Conducted cutting-edge research to develop assistive AI solutions, enhancing daily communication for the disability community through innovative machine learning applications.
Highlights
Optimized model performance for BISINDO sign language classification on Raspberry Pi 5, achieving a high-efficiency inference rate of 1.757 FPS (0.57 ms latency) through FP16 Quantization via TensorFlow Lite, validated by field testing metrics.
Engineered a Hybrid AI (RF + Transformer) architecture to decouple static and dynamic data processing, significantly boosting BISINDO sign language classification test accuracy to 98.57% and outperforming the LSTM baseline by 4.07% via rigorous benchmarking and real-time validation.
Designed and implemented a robust Computer Vision pipeline utilizing MediaPipe and OpenCV, enabling precise hand gesture feature extraction from visual inputs and optimizing image processing performance on edge devices.