About Me

CV (Current as of 12/2024): [Click to Download]
Google Scholar: Link
DBLP: Link

Hi, my name is Joel. I’m a Computer Science PhD candidate at the University of Waterloo, supervised by Dr. Lukasz Golab. My PhD research is focused on developing novel explainable artificial intelligence (XAI) methods for emerging deep learning models. I’m currently working to explain the behavior of retrieval-augmented LLM systems, i.e., pipelines that involve retrievers, LLMs (using RAG), and LLM agents.

We’ve published many papers that propose novel methodologies, algorithms, and tools to help explain (or generally to debug) each of these models. They can be organized into 3 larger projects (one for each model):

• CREDENCE: Explaining document rankers.
• RAGE: Explaining retrieval-augmented LLMs.
• LADYBUG: Debugging LLM agents.

The larger vision for my thesis is to leverage counterfactual reasoning to explain these systems’ outputs to laypeople. Counterfactual explanations help users convert undesirable outputs into desirable outputs (e.g., show how different data sources could lead an LLM to correct its incorrect answer). More generally, they also help users explore alternative outputs (e.g., show how different data sources could lead an LLM to different answers). We are the first to leverage counterfactual thinking for the purposes of explaining these emerging models, and will be the first to explain the sophisticated behavior of these larger systems. For more info, please see my Research Page or links above.

Before UWaterloo, I received my Bachelor of Computer Science (Honours) from the University of Windsor. Under Dr. Alioune Ngom, I researched and trained machine learning models for breast cancer diagnosis and treatment (drug repositioning). Under Dr. Robin Gras, I researched emerging language models (namely BERT, which was brand new at the time), fine-tuning said models to analyze the sentiment of social media posts.

Outside of UWaterloo, I’ve worked as a software engineer for a variety of clients. I currently lead a team of developers at Movyl Technologies, where we’re building data science tools at various stages of a proprietary AI / data pipeline. Prior to this, I worked as a software engineer for the Federal Government of Canada in Ottawa.