Projects Highlights & Completed Systems
1. Embrace Study (NDA)
August 2026
Description
Built secure participant-tracking infrastructure for a 200+ person prospective study on maternal-infant postpartum psychological adjustment. Architected the entire platform using Next.js, ShadCN, and Supabase, ensuring robust role-based access control, interpretable longitudinal insights, and strict medical-grade data protection.
(Source code and live deployment restricted under non-disclosure agreement).
2. Diabetic Retinopathy Screening (Champion @ NAIC 2026)
June 2026
Description
Developed a computer vision diagnostic pipeline for clinical retinal image analysis to detect diabetic retinopathy.
Preprocessing Pipeline: Implemented Contrast Limited Adaptive Histogram Equalization (CLAHE), Graham’s Method for local color normalization, and Feature Adaptive Cropping to isolate retinal regions and eliminate lighting variations using OpenCV and NumPy.
Model Architecture: Trained an ensemble architecture combining ResNet-34, EfficientNetB4, and a Random Forest meta-classifier, reaching 91% accuracy on limited, highly unbalanced clinical datasets.
Source code: Diabetic Retinopathy Screening GitHub
3. Byte of Kuih (1st Runner Up @ NAIC 2025)
July 2025
Description
Curated, augmented, and cleaned a proprietary dataset of over 16,000 images representing 8 traditional Malaysian kuih classes under diverse lighting, orientations, and backgrounds. Fine-tuned an EfficientNetV2 model using PyTorch and Hugging Face Transformers, reaching 98% validation accuracy and 100% test accuracy.
Read the full breakdown: National AI Competition 2025 Post
Model repository: NAIC Model 2025 GitHub
4. JotMe (Best Beginner Project @ LingHacks VI)
June 2025
Description
A mental-health digital journaling web application powered by NLP. Integrates KeyBERT for keyword extraction with fine-tuned emotion-detection transformer models, combined with a NetworkX graph-based contextual sentiment analysis algorithm to detect sarcasm and nuanced expressions, generating real-time crisis-intervention alerts and interactive ChartJS mood trends.
Read the hackathon story: LingHacks VI Blog Post
Source code: JotMe GitHub
5. Generative User Interface for Differential Evaluation via Symptom Analysis and Record Keeping (GUIDE-SHARK)
March 2025
Description
Designed to relieve hospital and emergency room overcrowding by giving patients a preliminary symptom screening before triage. Features a clean cross-platform Flutter frontend, coupled with a Django REST backend and PyTorch machine learning models to analyze multi-modal patient symptoms and suggest preliminary differential diagnoses.
Frontend Repository: GUIDE-SHARK Frontend
Additional / Earlier Projects
Artificial Intelligence with Reprompting & Code Execution: Iterative code generation and evaluation inspired by AlphaEvolve. GitHub
Maker Kehadiran: Custom MVC web application in PHP & MySQL for attendance tracking. GitHub
Study Scroll: Educational revision tool leveraging doom-scrolling micro-habits for A Level students. GitHub
