The development of Sign Language Technology (SLT) often stagnates at the computational prototype phase, creating a deployment bottleneck for practical implementation. This research aims to bridge this gap by integrating an existing machine learning model into a highly responsive web-based architecture for learning the Indonesian Sign Language (BISINDO). A Random Forest classification model, combined with MediaPipe for spatial landmark extraction, was engineered into a separated client-server architecture utilizing ReactJS and FastAPI. To achieve real-time inference with low latency, the WebSocket protocol was exclusively implemented for continuous video frame transmission and prediction feedback, decoupling it from standard HTTP operations. Functional testing and user trials were conducted empirically at SLB B YPLB Majalengka. The results demonstrated that the proposed architecture successfully maintained real-time detection capabilities with minimal client-side computational load. The integration provides an interactive self-assessment tool for deaf students, proving that lightweight computational models combined with optimized communication protocols can effectively transform theoretical algorithms into practical assistive educational technologies.
Copyrights © 2026