James Ridey

Software engineer

James Ridey

Software, systems,
and the work in between.

I’ve shipped software at Nullify, Canva, and Google. My work spans product engineering, performance investigations, and independent projects in inference, reverse engineering, and hardware.

View résumé PDF Contact me

Seeking relocation to the San Francisco Bay Area. Currently based in Sydney.

My JRB8 computer running on a Tiny Tapeout board
From an instruction set to silicon. My JRB8 computer, fabricated through Tiny Tapeout.

Skills, in context

Professional

Go, Python, TypeScript, JavaScript, React, AWS

AI & development tools

RAG interfaces, LLM-assisted development, Playwright, Storybook, ESLint

Independent projects

C/C++, CUDA, Verilog, C++ inference, STM32, ESP32, PCB design

Professional experience

Product engineering experience, with a particular interest in performance and difficult debugging problems.

Nullify

Product Engineer

March 2026–present

Performance and engineering tools for an AI application-security platform.

1.9 s 1.1 sGo Lambda initialisation
Development deployment
9.7 s 50 msPython startup overhead
Subsequent warm invocations

I removed Go startup reflection and reused a Python subprocess across warm invocations. These improvements targeted startup overhead, rather than representing end-to-end request latency.

Frontend rebuild & AI coding checks

I shipped a Vite frontend rebuild with an atomic architecture spanning approximately 470 components. After identifying recurring mistakes in generated code, I directed an LLM to implement ESLint rules for dependency direction, colour tokens, and shared primitives.

I also built opt-in visual checks using Storybook and Playwright. ImageMagick compares light and dark screenshots against a pull request’s base revision, with differences posted to GitHub for review. The checks are advisory.

Canva

Software Engineer

2020–2025

Making AI-generated designs practical to preview.

For Magic Design, I integrated generated text and selected photos into preset layouts outside the editor. Displaying many designs as complex DOM trees caused slowdowns. Working with editor stakeholders, I reused Canva’s canvas renderer to display image previews without retaining each design’s full DOM.

Team Brain: a working RAG prototype

I built the frontend for a prototype that answered questions using users’ documents, partnering directly with the backend engineer to deliver and demonstrate the end-to-end experience. This was a prototype, not a production launch.

Google

Software Engineer Intern

2017–2019 · Two summer internships

Adapting software to a 512 MB device.

I ported Neighbourly to KaiOS/JioPhone under a 512 MB RAM constraint and wrote the team’s first porting specification.

ThoughtDesign

Junior Software Engineer

2014–2017

Built web applications using ASP.NET MVC and Angular.

Independent engineering

Personal builds and open-source contributions.
From an unexpected result to a working system.

My e-paper art frame displaying a locally generated science-fiction scene
A personal system, generating new images every day.

C++ inference / Python / Raspberry Pi

Local AI, all the way to the display.

I wanted an e-paper art frame that generated its own images locally. Getting it working took me into model exports, tensor dimensions, tokenisation, and image-quality bugs.

I added SDXL Turbo support to OnnxStream’s C++ inference engine and contributed fixes to tokenisation, VAE scaling, and XNNPACK integration. Five pull requests were merged upstream.

The finished frame connects inference, a Python service, image history, web controls, and colour processing for the physical display. It sits on my shelf and runs every day.

SDXL Turbo contributionFrame source
Inside the debugging process

To investigate mismatched output, I decoded OnnxStream’s latents using the Diffusers VAE to help isolate the cause. Working with the maintainer, I corrected the export’s latent dimensions to resolve repeated image content. Later fixes addressed decoder scaling and CLIP tokenisation.

Read the investigation
JRB8 running on a Tiny Tapeout demonstration board, with its LEDs illuminated

Verilog / CPU architecture / Compilers

A computer from the logic up.

I designed an 8-bit CPU, instruction set, and assembler, tested it in simulation and on a Tang Nano 9K FPGA, then had it fabricated through Tiny Tapeout. I also built a language, compiler, and browser-based compile-flash-run tool.

Explore JRB8
Electronics from my Brother embroidery cartridge reverse-engineering project

Reverse engineering / Hardware interfaces

Making an old machine speak.

I built an Arduino reader to dump cartridge memory, investigated stitch data through hex and desktop-software analysis, and made a replacement cartridge the embroidery machine accepted. Others have reused the published findings and code.

Explore the cartridge project
The Minecraft pack.png landscape whose world seed was being searched for

CUDA / GPU profiling / Open source

Finding the bottleneck in a Minecraft seed search.

NVIDIA Visual Profiler exposed early thread exits. I split filtering and tree-position work into separate CUDA kernels, achieving roughly 2× performance on my machine. The contribution was merged upstream.

Read the merged contribution

There’s more on the workbench: robotics, electronics, and other experiments.

Visit my project workshop

Learning & credentials

Formal training, independent study,
and problems worth sticking with.

Flare-On reverse engineering
Completed every challenge in 2022–2025.
157th of 4,100+ entrants in 2025.
Burp Suite Certified Practitioner
Certified in 2026. Verify credential
Offensive IoT Exploitation
DEF CON training in firmware and hardware reverse engineering.
BEng (Honours), Software Design
Macquarie University, 2016–2020.

Let’s talk.

I’m interested in systems, performance, debugging, and hardware integration roles. If that sounds like your team, I’d like to talk.

Based in Sydney. Seeking relocation to the San Francisco Bay Area.

View résumé PDFDownload résuméLinkedInGitHub