HARDWARE–SOFTWARE CO-DESIGN ENGINEER

Engineering
across layers.

Hi, I’m Lucas (Lucas) Fang.
I connect the low-level with the big picture —
from GPU instructions to intelligent interfaces.

Explore my work View résumé

CURIOUS BY DEFAULT. BUILDER BY CHOICE.

FIG. 01 — A SYSTEMS PERSPECTIVE LIVE

The interesting things happen between the layers.

SCROLL TO EXPLORE ↓

PORTFOLIO / 2026

LOW-LEVEL THINKING✳HIGH-LEVEL POSSIBILITY✳BUILT TO WORK✳ALWAYS EXPLORING

01 / SELECTED WORK

Ideas made real.

Two projects. Different layers.
The same drive to understand and build.

OPENASSEMBLERSM_80 / AMPERE

OPENASSEMBLER / PTX → NATIVE SM80

gemm_fp16.ptx↗

// Capture a tensor tile and lower it.
.visible .entry gemm_fp16(
    .param .u64 A, B, C
) {
    mma.sync.aligned.m16n8k16
      row.col.f32.f16.f16.f32;
}

SM80 NATIVE BACKENDPTX → SASS

→→

ILLUSTRATIVE COMPILATION PATH · SELECT A STAGE

SOFTWARE THAT SPEAKS HARDWARE.↙ 001

01 / COMPILERS & GPU SYSTEMS

OpenAssembler.

Opening the black box
of GPU compilation.

A from-scratch PTX-to-SASS compiler for NVIDIA Ampere with native FP16/BF16 tensor GEMM, cooperative FP32 reductions, and explicit application workflows.

PythonPTX / SASSSM80 / FP16/BF16Compiler design

10–30%selective kernels¹

18/18reduction cases passed¹

¹ Repository-reported results; benchmark scope in case study.

GRAND TOOLAN IDEA → AN INTERFACE

✳

START WITH A LITTLE POSSIBILITY

01 Blueprint→02 Design→03 Review→04 Demo

A LITTLE CLARITY, AT A GLANCE

Good morning, builder.

Your workspace ↗

Total revenue$24,680↗ 12.8%

Active projects12+3 this month

A clearer view of your growthTHIS MONTH ↗

Try it — turn a prompt into a previewSIMULATED WALKTHROUGH

LESS FRICTION. MORE MAKING.↗ 002

02 / AI AGENTS & PRODUCT TOOLS

From prompt
to possibility.

An AI design workflow.
A shorter path to building.

Grand Tool brings language models, Pencil MCP, visual review, and code generation into one local workflow. Start with an idea and context. Leave with a reviewed design and an HTML or React demo.

AI agentsMCPPythonReact

01 Understand

02 Design

03 Review

04 Build

02 / THE PERSON BEHIND THE PROJECTS

At home between
disciplines.

I like understanding a system
deeply enough to make it better.

I’m Lucas Fang, a hardware–software co-design engineer with a background in applied mathematics and machine learning. My work moves between GPU compilers, decentralized learning, and AI-powered developer tools.

I’m drawn to the boundaries: where an algorithm meets an architecture, where a model becomes a working product, and where a good question turns into something you can run.

Explore my research

[01]

Close to the metal.

GPU architecture · PTX & SASS
Compiler backends · CUDA
Instruction scheduling · Verification

[02]

Built as a system.

Python · Systems engineering
Developer tools · MLOps
API design · Workflow orchestration

[03]

Intelligence, applied.

Machine learning · AI agents
Decentralized learning · MCP
Mathematical modeling

03 / A FEW STEPS ALONG THE WAY

Experiences.

The full résumé

JUN — AUG 2026

University of California, Santa Cruz

Machine Learning Intern

Investigated convergence in peer-to-peer decentralized learning networks. Built and integrated network simulations, designed experiments, and analyzed gossip learning behavior against a theoretical O(1/T) convergence rate.

Decentralized MLSimulationConvergence theory

GOOD SYSTEMS START WITH A CONVERSATION

04 / WHAT’S NEXT?

Let’s build
something interesting.

LucasFang2@outlook.com

GitHub ↗LinkedIn ↗Résumé ↗