Projects

GridFlow
RL and CV
January-March 2025
GridFlow is a smart traffic light system that optimizes traffic at a single intersection using reinforcement learning. The system employs YOLOv5 for real-time detection of vehicles and pedestrians. Q-learning and Deep Q-Network (DQN) algorithms dynamically adjust traffic light timings based on current conditions.

CoST: Code Switching TranslaDon Data AugmentaDon Method
NLP and ML
January-March 2025
CoST is a data augmentation method for code-switching translation tasks. It uses a fine-tuned mT5 model to generate synthetic code-switched sentences. The approach leverages machine translation to enhance the diversity and quality of training data. In evaluation, it more than doubled the accuracy of CoSDA-generated text on the XNLI benchmark.

AI-Powered H&E Visualization Tool
Full Stack Development
June-September 2024
Users upload H&E-stained histological images to be run thorugh a CV model which highlights potentially cancerous cells. The visualiation tool for the output includes zoom, scrolling, and a sliding window. It is deployed on AWS with a fully automated CI/CD pipeline, integrating services like Amplify, DynamoDB, S3, and SageMaker to manage hosting, the ML pipeline, and backend operations.
Last updated May 2025: the website is under construction by A*STAR in Singapore.

Battledart
Integrated Systems
March 2022
Battledart is an interactive Battleship game using C on a Raspberry Pi. Players throw magnetic darts at a custom-built 8x8 sensor grid, which detects hits in real time and displays the game on a computer monitor. The project involved configuring hardware, soldering connections, and integrating software to link the physical board with the digital game interface. All graphics were created from scratch.