The Malaysian driving education system is highly dependent on subjective manual assessments, leading to inconsistency in assessing crucial defensive driving manoeuvres such as blind-spot checking and inefficient workflows for booking practical exams. This project will identify training gaps and develop a predictive analytical architecture and prototype an AI-assisted web system to standardize the evaluation of JPJ practical tests.

The system was developed using the Agile Software Development Lifecycle (Agile SDLC). Structural requirements were transformed into normalized Third Normal Form (3NF) database schemas and iterative functional blueprints over four developmental sprints. We have developed a responsive web platform based on a dual-mode operation architecture: an instructor-supervised demerit portal and an unsupervised self-practice module using Google MediaPipe for real-time optical head-pose tracking (Yaw > 45°).

The system is powered by an asynchronous Python FastAPI backend which collates driving metrics to compute a machine learning test readiness score, programmatically intercepting exam bookings if a candidate’s score is below the mandatory 80% threshold. In Final Year Project I (FYP I), high-fidelity interactive web prototypes were successfully created and validated to fully visualize the core user workflows, vehicle inspection checklists and analytical dashboards. This free-of-charge infrastructure solution will eventually offer a standardized and data-driven framework to enhance candidate readiness and operational efficiency in Malaysian driving institutes.