AnakSehat

I designed a decoupled architecture where Next.js handled the frontend and user interaction, Express.js acted as the backend API and middleware, and a separate Flask service handled machine learning inference. This separation allowed the ML model to be developed, tested, and scaled independently.

I worked with a dataset of 121,000+ child growth records based on WHO Z-Score standards. I chose K-Nearest Neighbors (KNN) because the data was structured and numeric, KNN performs well on distance-based classification, and it provides high interpretability—important in health-related systems.

I implemented RESTful APIs using Express.js to handle authentication and user data, forward prediction requests to the ML service, and return results to the frontend in real time. This ensured clean communication between services and avoided performance bottlenecks.

Because the target users were parents and healthcare workers, I focused on minimal data input, clear classification outputs, visual result presentation, and chatbot support for guidance. This balanced technical accuracy with usability.

Given the tight deadline, I helped define clear API contracts early, modular task division across frontend, backend, and ML, and frontend development using mock data while ML was still in progress. This allowed parallel development without blocking the team.