Portfolio
I am an AI/ML engineer and researcher with an MSc in AI (Distinction) from the University of Bath and over four years of experience across academia and industry. I am specialising in reinforcement learning, computer vision, and statistical modelling, applying them to projects like generative modelling for stroke-ordered drawing and chemical recipe optimisation (e.g. for MXene).
My broader technical interests span MLOps, control and optimisation, reinforcement learning, quantitative analysis, natural language processing, computer vision, and computer graphics. This site showcases some examples of my end-to-end ML work and research, demonstrating my skills and interests across the diverse fields I am working on.
Writing
This case study explores `EdgeConv` and `DGCNN` to build an intuitive understanding of their working principles and implementation. It does so by using the ModelNet10 dataset for benchmarking on a 3D classification task and examines the mathematics behind the techniques. The article concludes with a reflection on the method’s capabilities and limitations, as well as some potential applications in cosmology and the arts.
A case study on using BERT and LIME to classify disaster tweets, interpret model behavior, and communicate both the strengths and limits of the system.
Projects
A compact view of applied machine learning, computer vision, and research-oriented work drawn from the broader portfolio.
A deep Q-learning agent trained in a custom environment to optimize MAX-to-MXene synthesis and enhance stability, leveraging a novel, custom-compiled dataset from the literature. The project will end with experimental validation of the agent's performance.
A PPO agent trained in a custom simulated environment to imitate in-between artists and draw line art from a reference and a character design.
Benchmarking AI techniques from existing literature to compare their baseline results. The project also is focusing on testing and adapting novel image fusion techniques derived from other fields to evaluate their direct application in satellite image processing.
Hands-on quantum machine learning work using Qiskit, covering classifiers, optimization methods, and applied experimentation with quantum circuits.