Hi, I'm James Yap.

An interactive slide deck that dives into the beautiful mathematics behind ReLU, in 3D space. Built using WebGL and Three.js.

A modern WebGL2-based viewer for point cloud data that applies polar unwrap on 3D LiDAR point clouds into 2D cylindrical projections with support for occlusion and Z-buffer depth testing.

CLI tool to automagically parse and compile YANG models into Typescript types with respect to Nokia SDN extended RESTCONF implementation. Lexer/parser grammar based on ANTLRv4.

Recreation of the 'Superliminal' game's 'Forced Perspective' effect using Three.js. Used Raycaster, Quaternion Rotations, Bounding Boxes, and Linear Transformations.

An interactive visualization of the heap data structure built with p5.js to teach an algorithms class at Carleton University as a Teaching Assistant.

Another interactive visualization of the Breadth-first Search (BFS) algorithm I built as a Teaching Assistant for a class of 700+ students. This time, it's the Snake Game!

A simple geometric demonstration to illustrate .attach() vs .add() in three.js

A conceptual video I made using 3b1b's open source math animation library (Manim), applied towards a Machine Learning background.

An Arduino-based wireless system that overrides door lock mechanism with a physical switch/servo in my dorm room. Works over the network from an iPhone app.

AirBoard is a portable digital whiteboard system engineered to allow teachers to make handwritten annotations on any flat surface using infrared diodes (~940nm), optical sensors, Computer Vision, and perspective transform.

ACTiON 2022 is the Annual Flagship Event of the 'Sunway Student Ambassadors' student leadership body. Landing page designed and built with WordPress.

Our project was a proof by concept that combines 3 different ML models: an LLM, a multimodal image processing model, and Higgs Audio V2, to create an immersive free-roam storytelling mystery puzzle inspired by Murder on the Orient Express by Agatha Christie. It won the 1st place award. (September 2025)

Our project 'VirtuDrip' (also known as Wardrobe AI) uses Snap AR's Camera Kit to create a virtual try-on fitting room that allows users to clone their friends apparel by converting 2D images to 3D wearable model with pose-tracking. It won the 1st place award. (17 - 19 May 2024)

Canada's biggest hackathon, hosted annually at the University of Waterloo. Our devpost link: https://devpost.com/software/sight-gonlt4 (15 Sep 2023 - 17 Sep 2023)

Our project 'BeGreen' uses Azure AI Vision to evaluate and assign individuals with a 'sustainability score' based on their social media activity. It won the 3rd place Overall Award. (1 March 2023 - 3 March 2023)

RecycleAI (also known as CleanVision) is what we created for HtH 2023. We built our project in the span of 48 hours and won the Makerjam @ Makercon 2nd place award. (3 March 2023 - 5 March 2023)

A graphical RegEx and automation workflow editor that uses Drag n' Drop with React Flow to create and visualize complex automation workflows. (1 Feb 2025 - 2 Feb 2025)

We created a stock-market trading frequency-and-size visualizer that takes after stars in the night sky with p5.js, a client-side library for creating graphic and interactive experiences. (21 Jan 2023 - 22 Jan 2023)

With the guidance of the Taylor Makerspace mentors and access to the Makerspace equipment, we were able to secure the 1st Place Award. (29-30 July, 5-6 August 2023)

Upon winning the competition, we were entered into Sunway iLabs's Startup Foundry program, which is a 6-month incubation program for startups. (6 Sep 2021 - 24 Sep 2021)

Automated classification of intraductal carcinoma of the prostate (IDC-P) using deep learning on non-linear multiphoton microscopy images. Explored transfer learning and CORAL (Consistent Rank Logits) to improve classification accuracy, with attention heatmaps highlighting the image regions most influential to the model's decisions.

Driver Risk Assessment using Dashcam Footage and Machine Learning. The DriveSafe model uses transfer learning on a pre-trained ResNet50 model to capture image features, and adds an Long Short-Term Memory layer (LSTM) to capture temporal data from video input.

Proof by concept puzzle game that combines 3 different ML models: an LLM, a multimodal image processing model, and Higgs Audio V2 to create an immersive free-roam universe inspired by Murder on the Orient Express by Agatha Christie. Utilizes PCM streaming over WebSockets for realtime audio generation and feedback, achieving ~1500 tokens/s throughput for audio generation, which translate to 60 seconds audio generation per second with higgs-audio-tokenizer.