Human-Centered Systems Builder

Jack Jiacheng Zhang

Signals Into Systems

I begin with a human problem, use mathematics to understand it, and build a system that can be tested in the real world.

Wearable sensing, applied mathematics, and embedded prototyping

Follow the signal from human need to working system

01 — Human Signals

Making hidden signals easier to act on

Fatigue can build before a rider recognizes it. A warning sound can fail when hearing is not available. These wearable projects begin with those gaps and ask how sensing can create a response that is immediate, physical, and clear.

EEG fatigue signal

Smart safety helmet · 16-channel EEG

The helmet project follows one chain end to end: electrodes held at a fixed position inside the shell, sixteen channels sampled from the forehead, and a warning that arrives as light and sound rather than as a notification. The shell, the sensor mount and the control enclosure were all drawn for this build, then printed and tested.

A test session: the helmet fitted with electrode wiring while a laptop displays live EEG waveforms.
Wired test session with live waveform monitoring.
CAD render of the helmet shell showing ventilation slots and the rear control enclosure.
Shell CAD — ventilation slots and rear enclosure.
The blue 3D-printed helmet shell resting on a work surface.
Printed shell, first fit check.
The transparent control enclosure holding the board and battery, next to the strap.
Sealed control box and strap.
  1. EEG16-channel acquisition
  2. ProcessingBLE to local computer
  3. AlertLED state and buzzer

Out of the lab

A unit was handed over to delivery riders at a donation ceremony. The people who would wear it were the ones who said whether it worked.

Delivery riders in green uniforms holding a fatigue detection equipment donation banner outside a building.
Donation ceremony with the riders who would wear the device.
A rider in a white helmet receives the fatigue detection device while the team member explains the equipment.
Handover and explanation at the ceremony.

Sound into touch

Sound-to-vibration wearable assistive device

A second prototype converts sound into a tactile response. Sensing and vibration sit in separate 3D-printed housings, so an alert can be felt in a noisy room, or by someone who cannot rely on hearing it at all. Trigger logic, response time and user testing are the next things to document.

Sensing and vibration modules in their printed housings.
Open housing with the board and battery.
  • Sound
  • Bluetooth
  • Vibration

A glove that carries the sensing

Wearable assistive device · in development

The third wearable moves the sensing into something a hand already knows how to wear. The enclosure is modelled so the electronics sit flat on the back of the hand and do not interfere with grip, then printed and fitted. The version shown is still being refined.

Wearing the glove prototype, with the sensor board and wiring mounted on the back of the hand.
First fit: board and wiring on the back of the hand.
Modelling the glove enclosure on a laptop, with the printed shell and components on the desk.
Enclosure modelling, with the printed parts alongside.
Visitors in cleanroom coats and shoe covers observing an electronics production line with stacked cartons.
Visit to an electronics manufacturing partner.

02 — Mathematical Models

Structure for problems too large to see at once

Some systems cannot be understood by examining a single object. These models trace relationships across energy, carbon, markets, and uncertainty — and every one of them needed a debugging story before it needed a result.

Diffusion factor models under ESG constraints

Co-authored with Zihao Yan · Beijing 101 High School

The paper asks how a diffusion factor model generates high-dimensional return scenarios, and how ESG constraints change what a portfolio is allowed to hold. The contribution statement credits me with studying the model's theoretical foundations, running and reproducing the baseline code, providing the computing environment, and writing parts of the results and discussion.

The theory was the easy half. The baseline code kept throwing errors and would not converge; the fix was a revised parameter initialization and more deliberate hyperparameter settings.

Title block of the paper: Diffusion Factor Models for ESG-Constrained Portfolios, by Zihao Yan and Jiacheng Zhang.
Paired histograms comparing synthetic and real distributions of maximum variance, minimum variance, maximum mean and minimum mean across six assets.
Synthetic against real distributions, four statistics across six assets.
  1. Scenarios
  2. Moments
  3. ESG constraints
  4. Attribution

Synthetic moments are estimated against real data first. Only once the scenario generator holds up is it constrained, and only then is performance attributed back to individual decisions.

Sustainable HPC

HiMCM Team 15204 · Problem B

The team modelled the electricity use and carbon footprint of high-performance computing through 2030, from manufacturing emissions to wastewater treatment. My contribution was the model framework and the extraction of the research data the model runs on.

  • 1,399 TWh Modelled HPC electricity use in 2030
  • −90.7% Carbon intensity, 347.5 → 32.2 gCO2/kWh
  • 873,600 t CO2 from water treatment at 15% reuse
Line chart of global HPC centre electricity use from 2010 to 2030, with original and predicted data.
Observed and forecast electricity use, 2010–2030.
Line chart of 2024 global CO2 emissions by HPC centre against the water purification ratio, rising from 5% to 45%.
Emissions against water purification ratio.

03 — Layered Prototypes

A working system is built across layers

The prototypes move across layers: a form that fits the body, electronics that stay stable in motion, code that carries a signal, and a test that reveals what should change next. An outer shell is not a finished system.

01

Form

The shell is drawn around the head first: profile, brow line, and ventilation slots placed where airflow actually reaches the scalp.

CAD render of the helmet shell in profile.

02

Fit

Electrode position has to survive being worn. Contacts are fixed to the interior with defined spacing, then checked again on a printed shell.

The printed helmet shell showing slot geometry.

03

Housing

A sealed enclosure holds the electronics, with a cable channel and a sliding lid so the unit can be opened without pulling on the wiring.

CAD render of the sliding-lid control enclosure.

04

Electronics

EEG front end, an ESP32-S3 for transport, and an N100 board for the work that is too heavy for a microcontroller.

The compact N100 computer used for local processing.

05

Signal

The wearable is not always in range. The BLE client rescans on a timer, reconnects when the device comes back, and forwards every notification to the serial port.

Code implementing BLE scanning, reconnection and a notification callback.

06

Feedback

A single fatigue value decides the output: past the threshold the LED goes red and the buzzer sounds, under it the LED stays green.

Code setting the LED colour from the received value.

Team 28660, FTC China Offseason

May 2025 · Judges’ Choice Award and Third Prize

A season of designing, building and testing a competition robot. The team took the Judges’ Choice Award and Third Prize, and my participation certificate is part of the record. My individual responsibility on the robot is still being written up.

Team members working on competition robot 28660 in the pit area during the build.
In the pit, mid-season build.
Judges’ Choice Award certificate for team 28660.
Participation certificate issued to Zhang Jiacheng of team 28660.

04 — Selected Recognition

External evidence, not the story

2025

  • S. T. Yau High School Science Award, Regional Second Prize
  • Conrad Challenge China, Silver Medal in Cyber-Technology & Security
  • FTC Judges’ Choice Award and Third Prize, Team 28660
  • Berkeley Math Tournament, Team Power Round National High Distinction
  • Berkeley Math Tournament, Team Overall National Distinction
  • AIME Qualification, Certificate of Excellence
  • MMATHS Team Round Distinction

2024

  • Australian Mathematics Competition, National Second Prize, Intermediate Level
  • Mini Berkeley Math Tournament, Gold in Team Overall, Team Round, Individual Round, and Puzzle Round
  • Mini Berkeley Math Tournament, Top 10 in Puzzle Round

I am most interested in problems where a hidden signal can become a useful system.

Holding the assembled controller board next to a laptop running the acquisition code.