Steven Chen's Website
HomeBlogProjectsPublicationsCV

CV

PDF
Name
Yucheng (Steven) Chen
Label
Machine Learning Research & Software Engineering
Email
stevenyc.chen@mail.utoronto.ca
Website
schen.me
Summary
Computer science student at the University of Toronto. I work on model reliability: uncertainty estimation for medical image segmentation with Project Neura / UTMIST, and a scaling audit of layer-local training that found Forward-Forward does not hold up on real data. Outside research I build production full-stack systems and my own options analytics tooling.

Work

Apr. 2025 – PresentResearch Intern
Project Neura / UTMIST
  • Research on medical image analysis, focused on uncertainty and reliability
  • Co-first-author on MIP Candy; co-author on SegWithU
Feb. 2025 – PresentAI / Full-Stack Engineer Intern
Kabuda Group, Toronto

Part-time. Kabuda runs a multi-service platform for newcomers and international students in the GTA.

  • Built conversational AI features and agent workflows for client-facing web and mobile products, and the REST backends they run on
2025 – PresentFounder
Amplimit
  • Home for my Forward-Forward scaling work and for Ampdraft

Education

Sept. 2026 – 2030 (expected)Toronto, Canada
Computer Science, Faculty of Arts & Science (University College)
University of Toronto
2025Ontario, Canada
Ontario Secondary School Diploma
Elton Academy

Publications

Apr. 2026
SegWithU: Uncertainty as Perturbation Energy for Single-Forward-Pass Risk-Aware Medical Image Segmentation
arXiv preprint; under review at UNSURE 2026 (MICCAI workshop) · Tianhao Fu, Austin Wang, Charles Chen, Roby Aldave-Garza, Yucheng Chen
Feb. 2026
MIP Candy: A Modular PyTorch Framework for Medical Image Processing
arXiv preprint · Tianhao Fu, Yucheng Chen

Technical Skills

Languages
Python, Java, C++, Rust, TypeScript
ML/DL
PyTorch, computer vision, medical image segmentation, uncertainty estimation
Web & Data
Next.js, React, NestJS, Prisma, PostgreSQL, Redis
Infrastructure
Docker, Nginx, Linux, Git

Projects

2026IV Surface Explorer

Options analytics dashboard, built for my own research

  • IV surface, smile, term structure, skew, GEX, open interest walls, max pain, expected move
  • Python/Dash pipeline that pulls and normalizes full option chains across every expiration
  • Exposed as an MCP server so a language model can query the pipeline directly
2026Ampdraft

Version control for writing

  • Snapshots, parallel variants, and word-level diff between any two versions
  • Deliberately avoids Git vocabulary; the model is presented as a timeline you can branch from
  • Next.js + Prisma + Postgres, with client-side PDF export via Typst compiled to WASM
2025ATLAS

Stock market pattern recognition with computer vision methods

  • Encoded price series as images (GASF/GADF/RP/MTF) and classified them
  • 83.4% validation accuracy with 17,081 parameters, 18.5x fewer than the ResNet-style CNN baseline
  • Sub-millisecond inference, validated across 200+ stocks
2025Portfolio Manager

Portfolio management and analysis tool

  • Built for the Wharton WGHS competition, Python backend with a Next.js frontend
  • Parallel pipeline supporting 60+ assets with sub-3-second response
  • SSE streaming for progressive metric loading