Projects List (In Progress / Ideation Stage)
- Double Pendulum Simulator and Balancer in Unity
AI-Powered Workout App- AI-Powered Doom Scroller for A Levels
- Task Assignment based on Availability
- Group Games VS AI
Projects List (Completed / Onhold for Ideas / Improvements)
1. Artificial Intelligence with Reprompting and Code Execution for Problem Solving
Description
The project takes in a problem statement, generates code to solve the problem, and then evaluates the code to determine if it meets the requirements. If the code does not meet the requirements, it generates a new prompt to improve the code. This process continues until the code meets the requirements or a maximum number of iterations is reached. This is an extremely simplified version of Google’s Alpha Evolve, which includes a component to improve the prompt based on the evaluation of the code.
Further Improvements
- Add more sophisticated evaluation metrics to determine if the code meets the requirements.
- Implement a more advanced prompt improvement strategy based on the evaluation results.
Source code here: AI with Reprompting and Code Execution
2. CNN Based Malaysian Traditional Cakes Image Classification
Description
This project is for a competition National AI Competition 2025, hosted by Hyperbyte. The task was to train a neural network to classify images of Malaysian traditional cakes. The dataset had to be collected by ourselves, and we managed to collect over 5000 images with a variety of lighting, colors and textures.
Initially, we started with Teachable Machine (TM) as a baseline, yielding a baseline accuracy of ~57% with 80 epochs and 0.0001 LR. In order to achieve a higher accuracy, we need to involve a more complex model compared to MobileNet utilized by TM. We switched to Keras (TensorFlow) using Google Colab and got an accuracy of around 80%. We applied some augmentations to our training images, such as color/contrast jitter, flips and rotations.
To improve its performance, … (will be released after competition ends)
Source code here: GitHub Link will be released after competition ends
3. Generative User Interface for Differential Evaluation via Symptom Analysis and Record Keeping
Description
This project is based on the issue of overcrowding in hospitals, especially Emergency Departments (ED). The goal is to create a generative user interface that allows patients to input their symptoms and receive a preliminary diagnosis, as well as a record of their symptoms for future reference. This can help reduce the workload of doctors and nurses in EDs, allowing them to focus on more critical cases. Research of this project is currently ongoing, and the project has an initial prototype powered by Google’s Gemini API. Backend code is currently private as more research is needed to improve the accuracy of the diagnosis.
Source code here: GUIDE-SHARK
Video for More Info
