Skip to content

Portfolio · Toronto-based · available worldwide

Cyrus Chan

I work on AI, machine learning, and natural language processing, mostly automated scoring.

[01]About

I like building things that improve people's lives. I care about not only technical feasibility, but also making an impact with good intentions.

With a background in computer science and language education, I see the world through both a technical and humanistic lens. I'm interested in projects that combine these perspectives, and I like working with people from different backgrounds.

[02]Experience

2025 — nowAI & Data EngineerVretta · Toronto
  • Build automated scoring engines for Canadian provincial K-12 assessments. Cut scoring effort by up to 50%.
  • Add safeguards and tools to the workflow to protect the integrity and privacy of 500k+ student responses a year.
  • Run research studies on automated scoring with education ministries and agencies.
2023 — 2025Full Stack DeveloperFreelance · Toronto
  • Built an open-source networking platform for Hong Kongers in Canada. Among its 5k+ users, 62% made connections, 23% got job referrals, and 6% got job offers.
  • Built a full-stack web app for a startup that helps young adults with ASD build romantic relationships.
  • Built websites for clients in several industries, with 10M+ visitors globally.

[03]Publications

[1]

Evaluating Machine Learning, Transformer, and LLM Approaches for Large-Scale Automated Essay Scoring

Cyrus Sai-Cheong Chan, Charles Anifowose, Su Zhang, Winnie Wing-Yee Tse, Nizam Radwan, Hyunah Kim

Proceedings of the Artificial Intelligence in Measurement and Education Conference (AIME-Con): Works in Progress 2026 · pp. 315–321

PDF

[04]Projects

Filter
TextropynlpwebLinguistic text analysis across three feature tiers.

An NLP pipeline that extracts 50 linguistic features from text. It analyzes one text or compares two.

Pipeline
Feature tiers
TierCountTypes
T139lexical, clause, sentence, punctuation, complexity
T27sentiment, coreference, cohesion
T34perplexity, surprisal

Deep link: #textropy · Live · Repo

DocChatllmwebAsk questions about your documents. Uses hybrid RAG.

A hybrid RAG pipeline with a ReAct agent. It answers questions from uploaded documents and falls back to web search.

Pipeline

Deep link: #docchat · Live · Repo

Automated Essay Scoring 2.0aesnlpKaggle essay scoring competition by The Learning Agency Lab, 2024.

Three attempts at the Kaggle competition, comparing ML and transformer approaches.

Results
ModelCV QWKTest QWK
RFR0.739130.75470
DeBERTa V30.801310.81713
DeBERTa V3 K-Fold0.803590.81222

Deep link: #aes2 · Kaggle

[05]Skills

languages
Python, TypeScript, SQL
ml
NumPy, scikit-learn, spaCy, PyTorch, Transformers, LangChain, LlamaIndex
web
FastAPI, Next.js, React, Tailwind CSS
tools
Docker, Git

[06]Education

2023 — 2025

Master of Science in Computer Science with Artificial Intelligence

University of York

2017 — 2022

Bachelor of Arts and Bachelor of Education in Language Education

The University of Hong Kong

[07]Contact