Portrait of Siyuan Spencer Liang

Siyuan (Spencer) Liang 梁思远

Hi, I’m an undergraduate at Cornell University studying CS and Math. I build software across web, backend, and XR, with research interests in Ubiquitous Computing and Physiological Computing.

At Cornell, I work with Prof. Cheng Zhang in the SciFi Lab, where I co-developed SonicFit, a smartwatch-only full-body pose tracking system. I’m also a BCI subteam member in Cornell XR, building EEG-based XR control systems.

Currently, I’m interning in the Pervasive HCI Group, working on extracting heart rate from treadmill-walking facial videos with Prof. Yuntao Wang.

Please feel free to contact me via sl3595@cornell.edu.

Skills

Programming Languages:
  • Python
  • Java
  • C#
  • C/C++
  • TypeScript
  • JavaScript
  • HTML/CSS
Frameworks & Libraries:
  • React
  • Next.js
  • Node.js
  • FastAPI
  • Spring Boot
  • Vite
  • PyTorch
Tools:
  • Git
  • Trellis
  • Docker
  • CUDA
  • PostgreSQL/PostGIS
  • Cloudflare
  • Render
  • Vercel
  • Unity
  • Linux

Experience

Jun 2026-Present

Beijing, China

Undergraduate Research Assistant

Pervasive HCI Group, Tsinghua University

Developed a motion-robust multimodal rPPG system that jointly learns from facial video, ring PPG, and IMU signals under ECG supervision to estimate heart rate during intensive exercise.

Sep 2025-May 2026

Ithaca, NY

Undergraduate Research Assistant

SciFi Lab, Cornell University

Co-developed SonicFit, a smartwatch-based full-body exercise pose tracking system that fuses IMU and active-acoustic echoes, achieving 8.61 +/- 0.818 cm MPJPE on a 3.6M-frame, 21-participant dataset.

Oct 2025-Present

Ithaca, NY

BCI Sub-Team Member

Cornell University Extended Reality Project Team

Contributed to a Muse EEG-based XR navigation project, collecting four-class movement-intent data and modifying 2 Unity XR environments for Meta Quest deployment and synchronized EEG data collection.

Publications

SonicFit: Cross-User Full-Body Pose Tracking for Fitness Exercises Using Active Acoustic and Inertial Sensing with a Single Commercial Smartwatch

Jiacheng Liu*, Saif Mahmud*, Siyuan Liang, Chi-Jung Lee, Ruidong Zhang, Cheng Zhang

IMWUT 2026 Under Review

Full-body pose tracking from a single smartwatch by fusing wrist IMU data with active-acoustic echo profiles, achieving 8.61 ± 0.818 cm MPJPE on a 3.6M-frame, 21-participant dataset.

Projects

AutoDL Server Monitor dashboard

AutoDL Server Monitor

Private-network monitoring dashboard for Linux GPU servers with live resource telemetry and read-only terminal visibility.

  • Collects cgroup CPU/memory, NVIDIA GPU, process, tmux, and GNU Screen data with explicit unavailable states.
  • Streams bounded terminal snapshots over WebSockets while keeping the API loopback-first and control-free.
  • Python
  • FastAPI
  • React
  • TypeScript
  • Vite
  • WebSockets
RouteCraft route planning interface

RouteCraft

Outdoor route planner for running, hiking, and cycling, with Strava import, GPX/KML uploads, and guided route generation powered by OpenStreetMap and a Spring Boot routing service.

  • Generates loop, out-and-back, and point-to-point routes within 8% of target distance.
  • Keeps spatial queries under 300 ms with PostgreSQL/PostGIS GiST indexes.
  • Next.js
  • React
  • TypeScript
  • Java
  • Spring Boot
  • Leaflet
  • PostgreSQL/PostGIS
  • OpenStreetMap
NailPilot application interface

NailPilot

AI nail try-on and merchant operations prototype serving 57 nail styles through a FastAPI backend.

  • Uses SAM3 segmentation, MediaPipe 21-point hand landmarks, finger-direction geometry, and alpha blending to place selected designs onto uploaded hand photos.
  • Ranks Xiaohongshu trends into merchant-facing operational actions from the dashboard.
  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • FastAPI
  • Python
  • PyTorch
  • OpenCV
BCI-VR Navigation Control demo

BCI-VR Navigation Control

A brain-computer interface project for controlling VR navigation with EEG signals.

  • Uses motor imagery for left and right movement and intentional blinks for selection in a Unity environment across 2 Meta Quest-ready XR scenes.
  • Reached about 61% movement accuracy and 88.7% blink-selection accuracy.
  • Python
  • Unity
  • C#
  • BrainFlow
  • MNE
  • scikit-learn
  • PyQt6