FeatureBoardby Lewis Valentine
Engineering portfolio

I build complex systems
across the full stack.

Real-time WebGL graphics, a from-scratch CAD geometry kernel, a CFD water solver validated against physics, computer-vision pipelines, edge AI, and the platform (FeatureBoard) that builds them.

Nine open-source projects, five that run live in your browser right now โ€” no install, no server. Everything below is real, tested code you can open and read.

9
Open-source projects
5
Live browser demos
0.1%
CFD error vs Torricelli
6
Domains, one stack
Flagship work

A few that show the range โ€” graphics, geometry, physics, and ML. Each links to a live demo or the source.

Gaussian Splats Live

My own 3D Gaussian Splatting renderer โ€” written from scratch, not gsplat.js or a ported viewer. Parses raw .ply captures with spherical-harmonic color and draws view-sorted gaussian quads on the GPU. Ships with a training/processing pipeline.

CAD Solver โ€” solid modeling kernel Live

A from-scratch JavaScript solid modeling kernel that delivers real variable-width edge blends โ€” the operation professional kernels were still fumbling as of 2025 โ€” across any dihedral angle and 3/4/5-edge corners, watertight by construction, out as standard STEP. The whole kernel runs in the browser: load the blended D&D dice-set benchmark from the viewer's Benchmarks menu, or drop in your own STEP/STL.

STL Wind Tunnel Validated

A CFD toolkit for STL meshes: watertight repair, volume meshing, and a flow solver. Its free-surface water model drains a tank and matches Torricelli's law to 0.1% โ€” validation baked into the test suite (39/39 pass), plus CAD import and automated boundary conditions.

Comet Rex Live

A 3D space runner in Three.js โ€” ride a dino-astronaut on a procedurally-detailed comet through selectable orbital trajectories, past photo-textured planets, with a real-time particle tail. Fully client-side, GPU-driven.

See all nine projects โ†’
Range

One engineer, one coherent stack, across domains most people specialize in just one of.

Real-time graphics

WebGL/Three.js โ€” games, GPU particle systems, a hand-written gaussian-splat renderer.

Geometry & CAD

STEP/STL parsing, surface tessellation, constrained Delaunay, watertight repair.

Physics & CFD

A flow solver with a free-surface water model validated against closed-form physics.

ML & computer vision

YOLO pipelines, a GPU video-montage engine, a Gemini auto-labeling + training console.

Edge & systems

On-device Gemini Nano (SlopRadar), a dependency-free Rust installer, Cloudflare deploys.

Product & platform

FeatureBoard itself โ€” the tool that plans, builds, tests, and ships all of the above.

How I build

Working with AI changes the good patterns. These show up across every project โ€” each grounded in real code on the How We Build page.

Radical provenanceEvery data load says where it came from and whether it worked.
Automate, don't askFix it in the pipeline instead of telling the user to run something.
Reduce to AI judgementGive context, ask for a verdict, get structured JSON back.
Validation over reviewEvery feature ships with an executable check โ€” the suite is the review.
Edge computeInfer where the data already lives; keep it on-device when it matters.
Planning as metadataFeatures start as structured plans that give the AI durable memory.

Let's talk.

Open to roles and collaborations. The fastest way to judge the work is to open a demo and read the source.

Lewis Valentine ยท built with FeatureBoard ยท Live demos ยท How I build