Research, writing, and life beyond the desk.
I’m Steven (Junyu) Chen, a student at The Storm King School. I love using my skills to help people and improve the communities around me.
My interests include environmental science, climate, sustainable development, and engineering. Away from my desk, I fence, play volleyball, and draw.
Research on climate policy, Parkinson’s disease, and building occupancy.
Research manuscript
Submitted to International Review of Public Administration · Under review
Submitted to the NYC Mayor’s Office of Climate and Environmental Justice · August 2026
New York City’s Local Law 97 limits emissions from large buildings. This paper examines how its rules, funding, and rent policies connect—and what evidence is needed to judge the effects on building owners and residents.
Peer-reviewed conference paper · Co-author
Published in Proceedings of the 2026 Design of Medical Devices Conference (ASME DMD 2026) · Paper DMD2026-1095
This paper compares compact AI models for predicting freezing of gait: sudden interruptions in walking that affect people with Parkinson’s disease. It tests how well the models work on sensor recordings from participants whose data they have not seen during training.
Conference paper · Co-author
Presented in person at the Duke Medical Robotics Symposium · October 2025
This study compares AI models that predict sudden interruptions in walking for people with Parkinson’s disease, using wearable motion sensors. The comparison weighs prediction accuracy, false alerts, and model size to inform future wearable assistance.
Conference paper · First author
Presented in person at the Northeast Robotics Colloquium, Cornell University · October 2025
This paper describes a soft, air-powered wearable suit designed to detect sudden interruptions in walking and provide a physical cue at the hip. It combines prototype hardware with a comparison of detection algorithms for Parkinson’s disease.
Research paper · First author
Submitted to ACM BuildSys · Decision pending
Status reported by Junyu Chen on October 30, 2026.
I developed this study with guidance from Prof. Chong Li, from preparing the data and comparing forecasting models to analyzing errors and writing the paper. In an offline test across 43 horizons, the selected hybrid model identified 14 sustained empty periods, with no recorded camera-label conflicts across 259 recommended 15-minute intervals.
The results use saved predictions from one building. Operational testing and measured energy savings require further evaluation.
Essay · AI & politeness
An essay on why a civil tone can help us communicate with AI, and where politeness becomes unnecessary.
Commendation · John Locke Institute Global Essay Prize · 2026
Poetry · Time & aging
Coffee rings on a table prompt a reflection on aging and the traces left by earlier generations.
Personal prose · Fencing & memory
A reflection on childhood fencing, a changing life, and the hope of returning to the sport.
Working draft
Descriptive prose · Forest observation
A description of a forest walk through scent, sound, light, and afternoon heat.
Working draft
Tools for meal planning, building occupancy forecasts, and school service hours.
I built meal-demand forecasts and staff dashboards to match meal preparation to attendance, with a reported reduction of about 60% in food waste. Typical service counts of 200 meals prepared for 150 guests imply an estimated 6,200 fewer surplus meals over two years, equivalent to about 3.1 metric tons of food.
Two-year projection assumes 2 services per week for 104 weeks and 0.5 kg per meal. Food-equivalent weight is estimated.
Programming language: Python · Dashboard tool: Streamlit
Explore the repositoryI built a public browser demo, offline Python replay kit, and notebook to help readers inspect an offline occupancy-forecasting study. The fixed hybrid rule found 14 sustained empty windows across 43 saved test horizons, with zero recorded camera-label conflicts across 259 recommended fifteen-minute intervals. Visitors can change the recommendation rule, share an example, download the replay, and recompute the comparison.
These are saved-output results from one building. Associated building-load overlap does not establish energy savings or future performance.
Python · Interactive browser replay · Reproduction notebook
I built a service-hours portal now used by 50 active users in The Storm King School’s National Honor Society, bringing hour tracking, teacher validation, and event sign-ups into one system. A two-stage review flow and teacher bulk-approval queue make submissions easier to verify, with approved and pending hours separated and review history retained. Capacity limits, sign-up deadlines, and waitlists help members and staff coordinate volunteer events.
Programming language: TypeScript · Web framework: Next.js
Explore the repositoryI help prepare and serve meals at a local soup kitchen. I also build forecasting tools to support meal planning.
I help collect food scraps on campus for composting and work with other students on recycling.
I help lead mindfulness sessions for low-income families in Newburgh and teach mindfulness and conversational English to children in China preparing for tumor treatment in the U.S.
A selection of my drawings and art projects.
Select an image to see it larger.
