01 · about
I’m Peter — a fourth-year student at Queen’s University double-majoring in AI and mathematics. My research lives where the two overlap: I evolve programs and study generative models, chasing AI whose reasoning a human can actually read. Right now that means working on evolutionary program synthesis for abstract reasoning in Dr. Ting Hu’s lab, with two papers in preparation.
Outside the lab I read a lot — English and Mandarin literature in roughly equal measure. I’m a Christian; when I noticed Christian students didn’t have a truth-or-dare game of their own, I built the first one.
I’ll play any sport you suggest — basketball, sprinting, badminton, tennis, swimming, squash — and I am, functionally, a pet person. Evidence below.
02 · research
interests: evolutionary computation · generative models & algorithms · interpretability · LLM reasoning
Evolving symbolic programs that discover abstract transformation rules on the ARC benchmark — reasoning that is transparent, auditable, and human-verifiable. Research Fellow with Dr. Ting Hu.
Studying the faithfulness of LLMs used as variation operators in evolutionary algorithms, probed through the game Contexto.
An interpretable clustering algorithm — groupings you can defend, not just plot.
Empirical study of 581 vulnerability lifecycles across 600+ repositories: 58.5% of fixes violated silent-fix protocols, with a median 33-day gap between release and disclosure. With Dr. Yuan Tian.
03 · education
B.Comp. (Hons.) — double major, Computing (AI) & Mathematics
Exchange — Informatics & Mathematics
04 · experience
Novel AI methods for abstract visual reasoning via evolutionary program synthesis, with end-to-end research ownership — design through evaluation, targeting peer-reviewed publication.
Supported 300+ students through tutorials, office hours, and (kind) grading.
Built the data pipeline behind a large-scale empirical study of security fixes in open source — scraping, parsing, and classifying metadata from 600+ GitHub repositories (Cohen’s κ ≥ 0.92).
05 · projects
Point it at your skincare — it reads 400+ ingredients and tells you the truth, in English or Chinese. Production RAG with ~5s end-to-end latency.
Astro · TypeScript · Supabase pgvector · OpenAI
The first online truth-or-dare game built specifically for Christian students — party game, fellowship edition.
HTML · CSS · JavaScript
Multi-agent pipeline that reads 80+ biomedical papers so a human doesn’t have to — semantic retrieval plus a T5 summarizer (ROUGE-1 0.48, ~98% less manual review).
Python · HuggingFace · ChromaDB · AWS S3
Talk to your terminal and it talks back — voice to shell command in under a second, with a safety-first intent classifier in between.
TypeScript · Node.js · ElevenLabs · WebSockets
06 · awards
07 · on the wall
Hung with clips, like they deserve.
08 · the pet department
Some are mine. Most just trust me quickly.
