Computer Engineering · Georgia Tech

Juefei Wang

I grew up in Shanghai and study computer engineering at Georgia Tech. I designed and patented a wearable that recognizes resistance-training exercises from accelerometer data on the device itself, wrote a first-author density functional theory study of magnetism in doped monolayers, and spent this past summer on object detection in aerial imagery at Shanghai Jiao Tong University. Outside of the engineering, most of my time goes to photography.

Published in RSC Advances · WiMo 2025

About

01

I have always been curious about electronics, and for a long time that curiosity ran ahead of anything I had been taught in a classroom. So I learned by building. The first project that went anywhere was a wearable for resistance training: an accelerometer, a classifier small enough to run on the device itself, and an app that could name the exercise, count the repetitions, and tell whether the form was falling apart. Two years later it had a patent and a conference paper behind it.

The habit carried into everything after it. I ran first-principles simulations of doped monolayers until there was enough there to publish as first author in RSC Advances, and then spent a summer on high-resolution aerial imagery, training detection models and testing how much of their performance actually came down to the way I had configured them. The subjects have almost nothing in common. The method does: set the experiment up carefully, then go looking for the place where the result does not hold.

What I want to work on now is the hardware itself and the software that sits closest to it: computer architecture and ML accelerators, on-device machine learning, and the perception systems that have to run on both. Outside of the engineering I do digital photography, mostly portraits and landscapes.

Program
B.S. Computer Engineering
School
Georgia Tech
Expected
2030
Interests
Architecture · On-device ML
Languages
Python, Java

Ask about my work

02
juefei@portfolio — ask.sh local index
Ask me about Juefei's research, the patent, the code, or the tools he uses. I answer from his actual record — papers, projects, and dates, nothing invented.

Built with a local answer index over my own record; where the viewer's Claude runtime is available, free-form questions are answered live by Claude from the same source material. It never answers from outside that.

Research & projects

03
2023 –
2025

RepSync — wearable resistance-training feedback

Independent project, in collaboration with SoftCom Lab, Cal Poly Pomona

Designed a low-cost wearable built around an accelerometer and an on-device ML classifier, paired with an iOS/Android app that identifies specific lifts, counts reps, and flags bad form in real time.

  • Granted a US patent for the device.
  • Co-authored the systems paper on its edge-CNN + Bluetooth LE architecture — presented at WiMo 2025.
  • Took First Place at GameGala, presenting to judges from Meta, Google, and NASA.
2025 –
present

Magnetism in doped Al₂B₂ / AlB₄ monolayers

National Graphene Research and Development Center

First-author DFT study asking whether substitutional doping (Mn, Fe, Cr) induces ferromagnetism in two lightweight, topological 2D semimetals — using ABINIT and Wannier90 for electronic-structure and spin–orbit-coupling analysis.

  • Found ferromagnetic ground states in 9 of 12 doped monolayers studied.
  • Showed Al-site substitution is thermodynamically favored over B-site in both materials.
  • Published first-author in RSC Advances (Royal Society of Chemistry).
2025

Group theory in crystallography

Pioneer Academics — mentored by Prof. Mark Tuckerman, NYU

Extended the monolayer work into a second paper applying group-theoretic symmetry indicators to crystalline structures, to predict topological properties directly from lattice symmetry.

Summer
2026

Aerial object detection for traffic monitoring

Shanghai Jiao Tong University

Ran a hyperparameter sensitivity study for YOLO-based detection on high-resolution aerial imagery, then paired trained models with classical SIFT features for multi-target detection of vehicles and pedestrians to flag traffic violations.

SOC gap opening · doped 2 × 2 supercell E_F E Γ K M Γ Δ dashed: without SOC · solid: with SOC Δ opens along M–Γ, just below E_F Defect formation energy by substitution site Al site B site Mn · Al₂B₂ Fe · Al₂B₂ Cr · Al₂B₂ Mn · AlB₄ Fe · AlB₄ Cr · AlB₄ 0 0.1 0.2 0.3 negative = stable formation energy (Ha)
Doped monolayers. Left, schematic: in the pristine primitive cells spin–orbit coupling produces no splitting at all, and only in the doped 2 × 2 supercells does it open gaps at the band crossings, here along the M–Γ leg slightly below the Fermi level, as reported for Mn substituted on a B site in Al₂B₂. Bands follow the Γ–K–M–Γ circuit used throughout the paper. Right, measured: defect formation energies from the paper's tables. Al-site substitution costs less than B-site for every dopant in both hosts, and is negative for all three dopants in AlB₄. Band panel redrawn for illustration; energies are the published values.
hyperparameter sweep configs varied, retrained aerial frame YOLO detector vehicles and pedestrians located violations flagged tiled boxes geometry SIFT features matched alongside the learned detections same frame
Shanghai Jiao Tong University. The detection pipeline. Aerial frames are tiled and passed through YOLO detectors retrained under systematically varied configurations, which is what the sensitivity study measures; classical SIFT features are matched on the same frames alongside the learned detections for multi-target localization of vehicles and pedestrians, and it is that geometry the violation rules are applied to. Drawn for this page.

Publications & patent

04
Journal article · first author

First-Principles Investigation of Transition-Metal Doped Al₂B₂ and AlB₄ Monolayers for Spintronics

Wang, J. F.; Luo, X. RSC Advances, 2026, 16, 1565–1584.
Conference paper · co-author

A Cost-Effective Wearable & Mobile System Delivering Real-Time Resistance-Training Feedback via Edge CNNs and Bluetooth LE

Wang, J.; Park, A. CS & IT-CSCP 2025 (17th Int'l Conf. on Wireless & Mobile Networks, WiMo 2025), pp. 189–199.
Patent · granted

Wearable Device for Anaerobic Exercise Tracking

US Patent — the hardware and classification method behind RepSync. Filing number available on request.

Skills

05
Programming & tools
PythonJavaPyTorch GitLaTeX
Computational methods
DFTABINITWannier90Linux / HPC
Also
iOS / Android devLightroom / Photoshop