3D artist.
Pipeline engineer.
Product builder.

Panoramic kitchen and interior configurators for some of the largest retailers in the UK, Europe, and North America. Clients include Wren, Howdens, and Starplan in the UK, Cosentino in Spain, and Toll Brothers and MasterBrand in the USA.

The consistent challenge across all of them: producing CGI-quality output at interactive framerates — photorealistic enough that customers feel confident in what they're buying before it's ever built.

My role spanned the full pipeline — designing the roomsets, building and managing the asset library, overseeing rendering and real-time output, and managing client relationships from brief through to delivery.

MasterBrand Cabinets needed a way to let customers explore their product range without visiting a physical showroom. The result was a browser-based virtual environment — a navigable kitchen space with selectable cabinet finishes and animated drawers, accessible on desktop, mobile, and VR headsets with no app download required.

The visual quality problem was an interesting one. Three.js had no reliable means of producing global illumination at the time, so we approached it as a CGI job — lighting and baking the entire scene in our render pipeline, then applying the result as lightmaps via a second UV channel. The geometry stayed light. The lighting stayed honest.

In 2019, a photorealistic, VR-capable experience that ran in a browser tab was genuinely new territory for the interior industry. The CGArchitect jury agreed.

Swoon needed photorealistic CGI images of a sofa across 120 fabric colourways. Doing that manually — rendering each variant individually, renaming files, compositing outputs — would have taken weeks.

Instead I built a pipeline that did it automatically. A Python script handled all render reference renaming inside a Fusion template, integrated with a Deadline render farm so that when each 3ds Max and Corona render finished, it triggered the script and a Fusion composite job. 120 final PNGs, hands-free.

The kind of problem that looks like a production task but is really an engineering one.

A browser-based sliding wardrobe configurator covering door configurations, glass and woodgrain finishes, internal layouts, and live pricing. Built once, licensed to multiple clients including Starplan, Bedrooms Plus, and Adamson — each with their own product range and branding on top of the same core tool.

The rendering quality was the hard problem. Real-time WebGL can't produce convincing soft shadows and ambient light natively, so we solved it the same way we solve it for CGI — baking the lighting to textures and applying them via a second UV channel. The result holds up at close range in a way that flat real-time shading doesn't.

My role went beyond the 3D and pipeline work. As the product evolved I became the primary point of contact for new clients — scoping requirements, managing onboarding, and making sure what gets built actually fits how they sell.

Wyrmwood make handcrafted hardwood desks for the gaming and creator market — a premium product with a passionate audience who care deeply about the details. The configurator had to match that energy. A standard product visualiser wasn't going to cut it.

The depth of customisation drove most of the technical decisions. Any component can be finished in any of the available hardwoods, managed through a painting-style UX that lets you apply materials directly to individual parts. Camera orbits and zooms are triggered by UI interactions, so the experience feels more like handling the object than browsing a website.

To hit the visual quality bar, I lit the 3D model in 3ds Max with Corona Renderer and used that setup to bake a custom HDRI environment map, which we then loaded into Three.js. The result was near-identical lighting between the CGI render targets I produced during the design phase and the live configurator — the kind of parity that's usually impossible between offline and real-time rendering.

Frovi's Bamboo shelving system is modular office furniture with an unusual constraint — the units are triangular prisms that must alternate in orientation for the shelves to sit correctly between them. Communicating that logic to a user, without making it feel like a logic puzzle, was the central design challenge.

The solution was a spatial planning grid where units snap into place in the correct alternating pattern automatically, freeing the user to focus on layout rather than geometry. The experience starts with curated presets — recognisable workspace configurations users can jump off from — then opens into a full 3D planner with accessories, frame colours, and an export that outputs both a PDF spec sheet and a downloadable 3D file of the finished design.

Frovi wanted something that felt different from a typical furniture configurator, and the UI reflects that — less product catalogue, more design tool.

Colony is Frovi's modular workspace pod system — privacy booths built from octagonal panel sections that can be configured as single units or linked together into larger multi-pod layouts, each fitted out with desks, shelving, screens, and accessories. The configurator needed to handle that structural complexity while staying approachable for a non-technical buyer.

The standout visual problem was shadow. Colony's panels are perforated metalwork — hexagon and segment felt cutouts, metal leaf — and the interplay of light through them onto the interior was a key part of the product's appeal, but exactly the kind of soft, complex shadowing that real-time WebGL struggles to fake convincingly. The solution was to prerender the shadows in V-Ray as opacity maps, projected onto trapezoidal planes shaped to match each panel's perspective, then have the dev team attach that asset to every exterior panel in the scene. The geometry stayed simple; the light did the work it needed to do offline.

The harder problem was underneath the visuals. Interior accessories follow a slot-based placement system — a desk can't go above a chair, certain combinations are structurally or spatially invalid — so the configurator had to enforce real layout rules rather than just letting users drop objects anywhere. That logic, combined with supporting both single and multi-pod layouts, also created a camera problem: the experience needed to shift perspective cleanly between viewing an exterior panel design and focusing in on an interior accessory layout, without ever feeling disorientating.

My role spanned the full front end of the project — client liaison and project management with Frovi, the configurator's design, and the 3D asset creation behind it. As with most configurators at this level of depth, the UI leans on presets first — a handful of considered starting layouts users can choose from before customising — to keep the cognitive load manageable given how many decisions the full configurator actually contains.

PGT's WinGuard range needed a configurator that could do more than show a window on a white background. The key selling point was glass tint — and tint only makes sense when you can see what it does to light coming through from outside.

The solution was to build two complete scenes — an interior living space and an exterior poolside view — rendered in real-time 3D and hidden from each other using occlusion planes that mask everything outside the window frame. As the camera orbits 180 degrees around the window, the parallax of the surrounding space shifts naturally with the viewing angle, and the scene transitions seamlessly from inside looking out to outside looking in. The glass tint options then read clearly against real light and sky rather than a neutral backdrop.

Simple in principle. The kind of thing that looks like it should have always existed.

Davea were building a furniture inspiration platform with an unusual technical ambition — a browser-based room planner that could deliver photorealistic renders on demand, not just real-time WebGL previews. The solution was a hybrid architecture: a lightweight Three.js front end for interaction, connected via WebSockets to an Unreal Engine instance running on Amazon S3, which handled the high-quality renders and streamed them back to the browser.

That meant building for two renderers simultaneously. The WebGL layer needed to be fast and responsive — optimised assets, simplified materials, low overhead. The Unreal environment needed proper lighting rigs, physically accurate materials, and a render pipeline capable of producing images that justified the wait. I used Rapid Compact to automate 3D asset optimisation, bringing per-asset production cost down from around £1,000 to £70.

The project won a Red Dot Award for Digital Solutions in 2022 — which had to be declined due to a client IP dispute. Davea subsequently folded. It never went to market, but as a technical case study it demonstrates something genuinely unusual: a working bridge between real-time WebGL and offline-quality Unreal rendering, accessible from a browser tab.

Threadmark were building a bespoke shirt platform — users could configure fit, collar, placket, sleeves, cuffs, pockets, buttons, and fabric across four shirt types. The challenge wasn't the UI, it was getting a shirt to look and behave like a real shirt in a browser, at a time when WebGL was still finding its feet.

Getting the fit right required a laser-scanned mannequin used as an avatar in Marvelous Designer, with each shirt rebuilt digitally from the manufacturer's own cut sheets. The edges of any interchangeable component were frozen so collars, cuffs, and plackets could swap without distortion. A baked lightmap was added via a second UV channel for realism — which led to the most technically interesting problem on the project: WebGL at the time had no standard format supporting dual UV channels, so we wrote our own, PikCTM, to make it work.

The client was venture capital funded and folded before it launched. The product was finished.

The Prophecy is Wyrmwood's flagship gaming table — a handcrafted piece of furniture that starts at five figures and goes considerably higher depending on wood species, seating, and accessories. At that price point, a real-time WebGL preview wasn't going to cut it. The visuals needed to match what customers were actually buying.

The solution was a fully prerendered configurator. Every component — frame, reversible play surface, topper, seating, accessories — was rendered in Corona Renderer as a masked layer, then composited in real time in the browser to reflect the user's exact specification. The result looks indistinguishable from a studio photograph because it essentially is one, assembled on the fly from hundreds of individually lit renders.

The UI went through several rounds of iteration to handle the sheer complexity of the product — exotic wood species priced individually, reversible surfaces with felt and wood sides, a full accessories catalogue including player desks, card organisers, and power packs. Getting that into something a customer could navigate confidently was as much of a design problem as a technical one.

RoomWiz is a browser-based flooring visualiser that manufacturers and retailers can license and embed on their own sites. Users upload a photo of their room or choose a preset, pick a flooring product, and see it rendered in place using a PBR pipeline that handles planks, herringbone, basket weave, and other lay patterns. CCT Elements and The Wooden Floor Store are the first clients live on the platform.

The harder problem was building the asset library. Sourcing product imagery meant scraping manufacturer websites, requesting files directly, and building a flatbed scanner rig for products that had no digital assets at all. Once source images were in hand, we needed to extract individual plank crops and generate full PBR texture sets — diffuse, height, normal, roughness — at production resolution. I built a Python GUI to handle the cropping workflow, significantly faster and more lightweight than Photoshop for this kind of repetitive task. It also integrated Meta's Segment Anything Model to attempt automatic plank extraction, though wood grain introduced enough noise that manual assist was still needed when plank edges weren't clearly defined.

The texture generation pipeline runs on ImageMagick and Python, pulling processing parameters from per-product data sheets that control how each map is generated. The system is also set up to export labelled dataset images for machine learning — the long-term goal being a fully automated pipeline once enough training data has accumulated.

Floorwiz is a browser-based epoxy and resin flooring visualiser — a sister product to RoomWiz, built for a completely different market. Where RoomWiz serves timber and LVT retailers, Floorwiz is aimed at epoxy suppliers and installers, primarily in the US. Torginol, one of the largest epoxy flooring suppliers in North America, were the first client and the brief that shaped the product. It has since been licensed across multiple US epoxy companies as a white-label SaaS platform.

The tool covers the three main epoxy floor types — metallic, flake/broadcast, and chip blend — each with its own customisation logic. Metallic floors are defined by pigment colours and swirl patterns. Flake floors are built from hex colour values controlling individual chip colours, size, and broadcast density. The configurator lets customers compose their own blends or choose from a library of presets, with the result previewed in a real-time 3D environment.

The technical centrepiece was convincing reflections in a browser-based, mobile-compatible renderer — a genuine constraint, since epoxy floors are essentially mirrors and the effect is central to the product's appeal. The solution came from early 3D game development, before raytracing and screen-space reflections were viable options: duplicating and mirroring the entire warehouse scene geometry below the floor plane, then rendering it with controlled opacity to simulate a reflection. No render targets, no expensive post-processing — just geometry and blending, fast enough to run on a phone. I built the 3D warehouse environment from scratch and configured it for Three.js, and handled the 2D background assets and metallic mask setup that drive the material variation across floor types.

This was the first commissioned interactive project I worked on at Pikcells — an educational tool for Johnson Controls' Sapphire fire suppression system, designed to help engineers and specifiers understand how the system works across different data centre installation sizes. Learning a complex engineering product from scratch and turning it into something navigable and visually engaging set the template for a lot of what came after.

The experience moves through several distinct environments — a building selection screen, an aerial floorplan showing how the suppression system connects across rooms, a detailed fire room, and a server room walkthrough. Each environment uses a different rendering approach suited to its purpose. The exterior buildings use a cell-shading technique borrowed from games — duplicate meshes with slightly inflated, inverted normals and a solid black material create crisp outlines that read clearly at small sizes. The fire room uses baked lighting and reflections applied to unlit materials for a clean, diagram-like quality. The server room is a prerendered equirectangular panorama, giving a high-fidelity immersive view without the overhead of real-time rendering.

Three rendering techniques in one project — each chosen for what it needed to do rather than out of habit.

Vita Student needed a way to show prospective tenants what life would look and feel like in their accommodation before moving in — and to let them browse actual available rooms at the same time. The result blended three distinct views into a single experience: a real-time 3D exterior of the building, an isometric floorplan showing live room availability by type and floor pulled from an API, and per-room views that toggled between a prerendered 360 panorama and a real-time 3D dollhouse model.

The toggle between Panoramic and 3D mode on each room was the detail that made it work. The panorama gave the feeling of being there — immersive, photorealistic, warm. The 3D view gave spatial understanding — where the bathroom sits relative to the kitchen, how the room is laid out. Neither alone was enough. Together they answered different questions a prospective tenant actually has.

Connecting the availability data to the visual experience meant the floorplan wasn't just decorative — rooms colour-coded by type updated in real time to show what was actually bookable. It turned a marketing tool into something genuinely useful.

Marshalls needed digital twins of their entire concrete and natural stone paving range — around 1,000 individual products — accurate enough to be used in AR, embedded on product pages, and fed into interactive configurators. The only way to do that at scale was to build a capture pipeline from scratch.

We built a three-camera photogrammetry rig on site at Marshalls, with a motorised turntable and controlled lighting to ensure consistent results across every product. Each paver went through the same capture process — photographed from multiple angles, processed in Reality Capture to generate a dense point cloud, then cleaned and optimised for web. I spent a significant portion of the project on site, either running captures myself, training and managing the team, or in meetings with Marshalls' digital marketing and product managers to keep the brief aligned as the library grew.

At that scale the logistics were as much of the challenge as the technical work. A thousand products means a thousand capture sessions, a thousand Reality Capture processes, a thousand optimised meshes. The assets ended up embedded directly on Marshalls' product pages and formed the foundation of their Outdoor Reality AR app — a tool that lets customers visualise paving products in their own garden using their phone.

Day & Knight needed a browser-based wardrobe planner that could handle real room dimensions — users input their actual wall measurements, place fitted cabinets across multiple walls, configure door styles, colours, handles, and internal storage layouts, then proceed to purchase. The finished tool is genuinely useful: clean, well-considered, and capable of handling the kind of complex L-shaped and corner installations that most configurators avoid.

I came into this project late, at a difficult stage. There were bugs, UX issues, and a client relationship under significant strain. The deployment sat on a third-party vendor's infrastructure who were largely unresponsive, which made debugging a particular kind of frustrating. I managed the client communications, worked through the logic and measurement issues, and navigated the deployment problems largely independently while the rest of the team was focused elsewhere.

It shipped. It's stable. Another company is now interested in licensing the same tool — which is probably the most straightforward measure of whether it ended up being worth the effort.

Marshalls' Modal paving range needed a way to let specifiers and landscape architects design a complete paving scheme before anything gets laid. The configurator lets users choose a laying course — herringbone, stretcher, chevron, basket weave, random — mix up to multiple granite colours with a ratio slider, select paver sizes, and see the result tiled across a full-scale 3D surface with a human figure for reference.

The UI was the main design challenge. Paving specification is a professional task with a lot of variables, and the tool had to work for both trade specifiers who know exactly what they want and homeowners who are exploring. The colour ratio mixer — sliders or percentage inputs to control how much of each granite appears in the random pattern — was the kind of detail that required several iterations to get feeling natural rather than technical.

The export options went further than most tools at the time. Users could download their design as a BIM file for Autodesk Revit, a PDF specification summary, or PBR texture maps for use in 3D visualisation software. Marshalls' own team still use it regularly for internal specification work — which is probably the most honest endorsement a tool can get.

Zephyr make premium kitchen ventilation hoods — the kind of product where fit, proportion, and finish matter as much as function. Before buying, customers needed to know how a hood would look above their specific hob, in their specific kitchen. The app let them do exactly that, placing true-to-scale 3D models of Zephyr's range into their own space using their phone camera.

The experience was guided — select a product, choose a size, set the installation height above the hob, then place it and explore. An explore mode let users orbit freely around the placed model with interactive hotspots highlighting key product features and specifications. At the time, AR on mobile was still early enough to feel genuinely impressive rather than routine.

My role covered the full pipeline — 3D asset production, scene setup, lighting and rendering management in Unity, UI design, and client liaison. Built in Unity for iOS and Android, it launched on both platforms in the same year.

An interactive cutaway visualiser for Steffes, a US heating manufacturer — built to explain the difference between their hydronic, forced air, and room unit heating systems. I modelled and textured the house, designed the UI, and handled the full visual build.

IQOS is Philip Morris International's heated tobacco platform — a global consumer electronics brand with the visual language and production standards to match. The brief came with a finished UI/UX specification and render targets: our job was to realise it in real time to the same standard, not to design it. That's a different kind of challenge to most configurator work, and an important one — reading a design spec and delivering it faithfully in a live WebGL environment requires a precise understanding of what makes a rendered image look the way it does, and how to close that gap in real time.

The configurator covers the full Iluma device range — Iluma, Iluma One, and Iluma Prime — each with their own colour options, door covers, rings, and full-body wraps in leather, crossdye, and silicon finishes. The customisation depth is significant, but the UI handles it cleanly, guided by the client's own design system.

The animations were the part I was most focused on getting right. Device components — the pen, the case, the door cover — fly in, rotate into position, and seat themselves as selections are made. The camera moves independently, pushing into detail or pulling back to show the full device in context. None of it is scripted in a way that feels mechanical — the timing and easing were tuned carefully to feel like the product is being handled rather than switched. At that level of visual finish, the difference between good animation and great animation is almost entirely in the easing curves.