Event language
UI language
Chronic diseases remain one of the world’s greatest health burdens. According to the <strong>World Health Organization</strong>, <strong>830 million people</strong> are living with <strong>diabetes</strong> globally, <strong>1.4 billion adults</strong> have <strong>hypertension</strong>, and <strong>high cholesterol contributes to 3.6 million deaths</strong> every year (<strong>World Heart Federation</strong>). These conditions are largely <strong>preventable</strong>, yet <strong>early detection</strong> remains a major challenge. In this conference session, I share the <strong>end-to-end journey of building HealthPredictor AI</strong>, an <strong>applied AI/ML health-prediction system</strong> developed at <strong>NUS</strong> using <strong>open-source tools</strong>, designed to predict the risk of <strong>chronic conditions</strong> such as <strong>diabetes, hypertension, and hyperlipidaemia</strong>. Our mission was simple: to <strong>empower individuals and caregivers with predictive insights</strong> for <strong>early detection and intervention</strong>, helping reduce the global burden of these diseases. Our models achieved an <strong>average accuracy of 98.2%</strong>, and we implemented a <strong>rule-based recommendation system</strong> to provide <strong>general follow-up suggestions</strong> and <strong>lifestyle guidance</strong>. We will walk through the <strong>practical engineering workflow</strong>: <strong>data cleaning, feature engineering, model selection, evaluation metrics, architectural decisions</strong>, and how we integrated both a <strong>lightweight RAG-based AI assistant</strong> and a <strong>simple rule-based recommendation system</strong> into the application to support <strong>real-time responses</strong> and <strong>contextual guidance</strong>. The session highlights <strong>real-world constraints</strong> – limited resources, limited data, and a small software engineering team – and how we successfully navigated them to build a <strong>clinically inspired ML system</strong>. Beyond the technical pipeline, I will share <strong>practical startup lessons</strong> from the project’s <strong>acceptance into the Microsoft for Startups Founders Hub</strong> – including what worked, what failed, and what we would do differently if rebuilding the system today. These include <strong>product thinking, user insights, data challenges, and engineering trade-offs</strong> that shaped the system’s development. This talk is designed for <strong>developers, students, and engineers</strong> seeking <strong>practical, implementation-ready guidance</strong> on building <strong>applied AI/ML systems using open-source tools</strong> – without requiring <strong>massive infrastructure</strong> or a <strong>research background</strong>. <strong>Conference Video Link:</strong> <a href="https://www.youtube.com/live/OoU3GWdNKPQ?t=24266s">https://www.youtube.com/live/OoU3GWdNKPQ?t=24266s</a>
