Projects worth bookmarking.

As part of the Cohere 4 AI community, trained a novel Vision Language Model on a multilingual instruction dataset curated by the team. Published paper on arXiv.

Contributed notebooks and documentation for knowledge distillation in computer vision and PII detection for LLM gateways as part of community initiatives.

A distributed chess system that runs multiple chess agents on morph cloud instances with real-time visualization.

Released a pip package to automatically label text, image, and video data using LLMs for topic identification. Supports local LLMs via Ollama and pydantic for structured output.

Led a University of Waterloo Data Science Club team to build a labeling tool, create custom datasets, and fine-tune DistilBERT for climate disclosures—unlocking 16 new signals and slashing LLM API costs by ~99.9%.

Fine-tuned a Llama-based model on synthetic data to answer natural language queries about an SQLite database by generating SQL and interpreting the results. Achieved 86% accuracy on held-out tasks.

Trained a YOLO model for enemy detection and used the detections to prompt Cohere’s Command-R+ for movement and combat actions.

A Dreambooth fine-tune of Stable Diffusion on a very important subject: my dog, Bella.

Created synthetic data with simulated drift for an introductory lesson covering TensorFlow Extended, drift monitoring, and CTGAN-based tabular generation.