Oliver Kwun-Morfitt

Learning to build ML at scale.

Building out ELT pipeline, improving pricing model accuracy, and improving backend performance.

Building CLI tools allowing remote access to compute.

Building scalable full-stack web application for the club's community.

Exploring predictive modeling techniques in trading applications. Published to deMistify.

ML experimentation infrastructure for Named Entity Recognition on job postings. Benchmarked 8+ models with automated eval harness measuring F1, per-class precision/recall, overfitting gap, inference latency, and confusion patterns. Achieved 51% F1 with distilbert-base-uncased at 3.47ms inference latency.

An open-source notetaking application that uses LLMs to reduce organizational overhead. Features include automated note connections, topic clustering, and intelligent content parsing—all while maintaining a minimalist interface where AI augments the experience in the background.

Quill is a SQL query optimizer that leverages ReST^EM to deliver frontier model performance on query optimization tasks. By integrating LLMs into the optimization with QWEN-7B, Quill achieves significant latency reductions while maintaining accuracy.

Parsec is a Python SDK that ensures LLM outputs conform to specified structures using Pydantic models. It simplifies prompt engineering by generating structured prompts and validating responses, enhancing reliability in applications like chatbots and data extraction.

Minimal double-entry ledger API in Go with balance invariants and a pluggable store. The current build ships an in-memory store and REST endpoints for accounts, balances, journals, and entries.

A swipe-based discovery platform for students built at UofT Hacks 13. Uses AI-powered embeddings to personalize recommendations as users explore opportunities across two modes: Matcha Mode for hobbies and Coffee Mode for career interests.