Zhizhou He

AI Infrastructure · ML Systems · Performance

Engineering
the AI
stack.

Stay curious. Build something. Make it better.

I’m Zhizhou He, a software engineer at Huawei working across AI platform engineering, scientific Python performance, and native build infrastructure. My recent work includes NAIE/Federated workflows for model retraining and evaluation, while my personal work extends into language-model training, inference runtimes, and reliable AI systems.

Explore my work Get in touch

Huawei AI platform · ML systems · Kunpeng

MODEL SYSTEMSPERFORMANCEBUILD & RELEASE

01 / SELECTED ENGINEERING

Behind the model.
Inside the system.

From model runtimes to enterprise AI workflows, scientific computing, and native delivery. A practical view of the infrastructure around AI systems.

PERSONAL SYSTEMS PROJECTCPU REFERENCE

miniGPT

From a Transformer
to a serving runtime.

Built a PyTorch GPT system spanning exact-resume training, continuous batching, paged KV management, prefix caching, and HTTP/SSE serving—with reproducible benchmarks and release checks.

  • PyTorch
  • KV cache
  • Continuous batching
  • HTTP / SSE
Explore the system

PROFESSIONAL WORK / HUAWEI · NAIE

AI platform.
Model lifecycle engineering.

Work on NAIE/Federated workflows around optical-fiber sensing: dataset path resolution, retraining and evaluation task assembly, online-learning device flows, and model-package delivery.

  • NAIE / Federated
  • Java / Spring
  • Model lifecycle
Read the AI platform case

PROFESSIONAL WORK / HUAWEI

Scientific computing.
Performance engineering.

Analyze NumPy and pandas performance on Kunpeng, review downstream optimization migrations, and validate changes against correctness and workload-level regressions.

  • NumPy / pandas
  • ARM64
  • ASV
Read the engineering case

PROFESSIONAL WORK / BUILD SYSTEMS

Native compute.
Repeatable delivery.

Contribute to migrating a C++/Python computing library from CMake to Bazel, packaging native extensions, and comparing build artifacts and test behavior.

  • Bazel / CMake
  • pybind11
  • Packaging
Explore the build pipeline
MORE ENGINEERINGData tools, systems foundations & earlier work

PERSONAL PROJECT / IN DEVELOPMENT

Stock Tracker

A research-tooling prototype focused on decision interfaces, point-in-time data integrity, and visible failure states.

Read the project brief

2022 / C++ · QT · MULTITHREADING

MIPS Simulator

An instruction-level simulator with a lexer, parser, virtual memory, and a thread-safe queue between the simulation and UI.

Project source (opens in a new tab)

2024 / C++17 · VORONOI · GEOMETRY

CenterlineGen

An autonomous-driving lane-centerline pipeline built around point Voronoi geometry, deterministic C++ graph extraction, quality filtering, and Python-based validation.

Project source (opens in a new tab)

2022 / PYTHON · OPENCV · PYQT

Drive S.A.F.E.

A two-person hackathon project for real-time fatigue detection and a calibration interface. Awarded overall second place.

View on Devpost (opens in a new tab)

02 / PROFESSIONAL EXPERIENCE

Compute.
Infrastructure.
Engineering depth.

My professional foundation spans AI platform engineering, scientific computing, native systems, and algorithmic geometry—building systems that can be measured, tested, and explained.

Professional contributions below; personal model systems are presented separately.

Huawei

CURRENT FOCUS

AI platform, scientific computing & build infrastructure

NAIE / Federated · Kunpeng · Java / Python / C++

  • Work on NAIE/Federated service flows for optical-fiber sensing, including retraining/evaluation workflows and device-side online-learning interactions.
  • Implement and debug Java/Spring task assembly around dataset path resolution, evaluation datasets, model versions, result paths, and request serialization.
  • Trace service failures across request routing, API conventions, mapping, and serialization; distinguish interface-specific behavior rather than treating all gateway calls as interchangeable.
  • Work on model deployment flows that locate a versioned model package, extract model.zip, and dispatch the package through the online-learning/device delivery path.
  • Continue scientific-computing and build work on the Kunpeng Python ecosystem: NumPy/pandas performance analysis, Bazel packaging, native dependencies, and validation.
AI platform engineering case

Huawei

EARLIER ROLE

Silicon validation & test automation

Charging systems · Hardware–software integration

  • Developed charger-compatibility automation and supported silicon bring-up, system testing, and long-running stability investigations.
  • Worked on test-equipment PCB development, board-level verification, and power-device reliability analysis.

GAC R&D Center

JUN — AUG 2024

Autonomous Driving Algorithm Engineer Intern

Autonomous driving · Computational geometry · C++ / Python

  • Built a lane-centerline generation pipeline from ordered lane boundaries using a point-Voronoi kernel, geometric filtering, endpoint normalization, and deterministic graph/path extraction.
  • Refactored the prototype into a modular C++17 core and CLI, then added quality validation including boundary densification, continuous boundary-distance checks, source-aware path splitting, and optional road-region constraints.
  • Developed Python visualization and evaluation tooling, synthetic geometry checks, and real-data replay; added CMake-based Linux/Windows CI plus ASan/UBSan coverage.
  • Validated the pipeline on a 371-frame replay dataset covering 2,777 lane boundaries and 258,291 input points, preserving all records while checking structural graph invariants.
CenterlineGen project (opens in a new tab)

SuperRobotics

MAY — AUG 2022

Embedded Software Engineer Intern

Developed embedded C software on STM32MP157D for an automated patrol robot, integrated sensors, and contributed verification plans and test reports.

03 / ENGINEERING PROFILE

At the intersection
of models
and systems.

My focus is AI infrastructure and ML systems.

I work across model lifecycle infrastructure and lower-level systems: enterprise AI workflows at Huawei, scientific-computing performance, native build/release paths, and a personal language-model runtime.

The common thread is turning a working prototype into a system with explicit state, controlled resource ownership, reproducible measurements, and a verifiable release path.

Code & technical projects (opens in a new tab)

Technical focus

Core engineering
Python · C++ · C · Linux · Git
Training & inference · personal systems work
PyTorch · Transformer · Exact resume · KV cache · Continuous batching · HTTP / SSE
Performance · professional work
NumPy · pandas · ARM64 · ASV · CPU profiling · Regression analysis
AI platform & model lifecycle
NAIE · Federated · AIOps · Java / Spring · Dataset & evaluation task assembly · Model deployment
Build & delivery
Bazel · CMake · pybind11 · Native linking · Python packaging · Automated testing

Education

Carnegie Mellon University

Graduate study · Electrical & Computer Engineering

Virginia Tech

B.S. · Computer Engineering

Earlier work includes bio-inspired robotics research, embedded-programming teaching, and C++ systems projects at Virginia Tech.

04 / CONNECT

Building the systems
behind AI?

Let’s connect about AI infrastructure, language-model runtimes, performance engineering, and developer tools.

GitHub (opens in a new tab)