Artlist Studio
Contributed a lot to Artlist Studio, especially its generative workflows and infrastructure, and a number of substantial UX features.
About
Over twenty years of pixels.
I studied Computing Science at the University of East Anglia from 2004 to 2007 and graduated with a first. I stayed on for a PhD, funded by a scholarship. My research was about rendering crowds on the GPU.
In 2010 I joined FXhome. For over a decade I was part of the small team that built its video and image software. I worked on the graphics side: effects, plugins, rendering and UI.
Since Artlist acquired FXhome, my work has moved to the full stack, focused on generative model workflows: the pipelines that turn models into reliable products.
I also volunteer as a coach at codebar, helping people from underrepresented groups learn to code.
Work
What I built
Artlist AI Toolkit
Improving the user experience and reducing generative error rates, to raise the conversion rate.
Artlist MCP
Worked on implementing the Artlist MCP server, which lets AI assistants such as Claude generate images, video and voiceovers from an Artlist account.
Artlist and Motion Array plugins
Worked on the extensions and plugins that bring Artlist and Motion Array into editing software such as Premiere Pro, After Effects, DaVinci Resolve and Final Cut Pro.
HitFilm
Joined to work on HitFilm 1. Implemented most of its visual effects across every version, and the hosting of third-party plugins. Also a significant part of everything else in the app, including the viewer interactions and all text rendering, with our own implementation of the Unicode algorithms for multi-language support. Most of our apps used Qt for the interface.
Ignite plugins
Implemented almost all of the effect plugins in C++, with the graphics in OpenGL. They run natively on several host SDKs, including Adobe AEX and OFX; FXhome was part of the OFX consortium.
Imerge Pro
Significant contributions to the product, whose effects mainly ran on OpenCL.
PhotoKey
Already existing when I joined. I substantially improved effect performance and quality, with many other improvements over the years.
CamTrackAR
Wrote the green-screen algorithm and made it run efficiently in real time on phones, using Metal.
Next-generation video engine
Heavy involvement in a rendering engine built from the ground up for efficient video and effects processing, using DirectX on Windows and Metal on Mac. To exercise it I built internal testing video editors on top, in WinUI 3 and SwiftUI. It supported fully ACES-compliant colour-managed video editing workflows. It worked very well, and I am proud of it.
Timeline
- 2010
Joined FXhome to work on HitFilm 1
Hired into a small team of developers to build HitFilm, released in July 2011. From then on I worked on every FXhome product. [Pro Video Coalition]
- 2010–2021
Effects, plugins and rendering across the product line
Implemented most of the visual effects in every app, and virtually all of the Ignite plugins. HitFilm alone reached over 2.4 million users worldwide. [Pro Video Coalition]
- 2021
FXhome acquired by Artlist
Artlist acquired FXhome in July 2021 and the software team joined Artlist. FXhome reported a community of over 6 million users at the time. [Pro Video Coalition]
- Now
Full stack at Artlist
My role shifted to full stack, specialising in generative model workflows, with Temporal for orchestration.
Research
Published papers
Real-time crowd rendering on the GPU, with Stephen Laycock and Andy Day at the University of East Anglia.
- GPU Techniques for Creating Visually Diverse Crowds in Real-Time
ACM Symposium on Virtual Reality Software and Technology (VRST), Bordeaux, 2008 - Illuminating Crowds Via Integral Approximation of Cosine Lobes
Computer Graphics International (CGI), Singapore, 2010 - Dynamic Animation of Large Crowds
Computer Animation and Social Agents (CASA), 2010
How I work
Give me the problem nobody else knows how to solve.
When a new task lands that no one has done before, I'm usually the one who gets it. I work out how it's done and make it work well.
- Made a green-screen algorithm run in real time on phones.
- Wrote a text-rendering engine with our own Unicode algorithms for multi-language support.
- Built a video rendering engine from scratch for DirectX and Metal.
- Moved from GPU graphics to full-stack generative model pipelines.
- Worked on more things and technology stacks than I can remember.
Skills
Core tools
A selection. I pick up whatever the problem needs.
C++PHPOpenCLMetal Shading LanguageQtColour theoryHTMLAWSDatadogMLACodexImage modelsBlack Forest LabsC++PHPOpenCLMetal Shading LanguageQtColour theoryHTMLAWSDatadogMLACodexImage modelsBlack Forest Labs
C#SQLVulkanGPU renderingSwiftUIAdobe AEXCSSCloudflareMixpanelDSAOllamaGoogleLightricksC#SQLVulkanGPU renderingSwiftUIAdobe AEXCSSCloudflareMixpanelDSAOllamaGoogleLightricks
PythonPrologDirectXCompositingSwiftPremiere ProReactFirebaseJiraCSAGitHub CopilotByteDancePythonPrologDirectXCompositingSwiftPremiere ProReactFirebaseJiraCSAGitHub CopilotByteDance
TypeScriptBashMetalVisual effectsWinUI 3After EffectsSvelteArgo CDTransformersEngramMCPOpenAITypeScriptBashMetalVisual effectsWinUI 3After EffectsSvelteArgo CDTransformersEngramMCPOpenAI
JavaScriptShellCUDAText renderingACESOFXNext.jsDockerMHAMixture of expertsUXKlingJavaScriptShellCUDAText renderingACESOFXNext.jsDockerMHAMixture of expertsUXKling
RustPowerShellGLSLUnicodeColour managementARKitNode.jsGitMQAKV cacheGenerative model workflowsMiniMaxRustPowerShellGLSLUnicodeColour managementARKitNode.jsGitMQAKV cacheGenerative model workflowsMiniMax
JavaOpenGLHLSLCrowd renderingColour sciencePlugin hostingTemporalGitLab CIGQAClaudeVideo modelsAlibaba WanJavaOpenGLHLSLCrowd renderingColour sciencePlugin hostingTemporalGitLab CIGQAClaudeVideo modelsAlibaba Wan
Talks
Presentations
A Deep Seek into LLM Architecture
How large language models work under the hood, from a single neuron to DeepSeek's attention and mixture-of-experts designs.
Slides coming soonContact
Let's talk.
Interested in my work, a talk or a collaboration? The best way to reach me is on LinkedIn.