RACKLIGHT: a 3D warehouse operating system you can click through, built as a concept

VITON¹³ builds concept products to show range where we have no client project to point to yet. RACKLIGHT is the operations one: a WMS for Norvane DC-02, an invented third-party-logistics warehouse with 400 products, 38 outbound orders a day, five inbound and five outbound dock doors and a cold room. The company, its customers, carriers, suppliers and every number are fictional, and the demo says so in a fixed badge on screen.

The brief we set ourselves: warehouse software usually shows stock as tables of codes, so make the warehouse itself the interface, with every pallet visible and every action reflected on the floor, without turning it into a 3D showpiece. It had to carry workflows a shift lead would recognise, show the numbers behind every suggestion and work in English and Russian, on a laptop and on a phone.

The warehouse is a 108 × 80 m building drawn in three.js with an orthographic, isometric camera: 12 main aisles with a pick face and bulk reserve, 4 aisles in a glass cold room, receiving, packing cells with a conveyor, shipping lanes and 10 dock doors with trucks. The 1,656 bin locations are instanced pallets coloured by stock level, or by pick frequency in heat mode. Nine pickers and five forklifts move along the aisle graph, trucks reverse onto the docks on schedule, and hovering a bin shows its SKU, quantity and fill.

Four tools work on it. Search (⌘K) covers the 400 SKUs, orders and bin codes; it flies the camera to the bin, lights it with a beam and draws a path from the nearest picker. The pick-path optimiser compares the order sheet’s sequence with a route built by nearest neighbour and 2-opt on shortest aisle distances, then sends a picker to walk it. The inbound scan’s putaway rule picks a slot by velocity, with fast movers nearest packing, slow movers on the upper reserve levels and chilled goods in the cold room, while a forklift carries the pallet there. Inventory intelligence is the fourth.

Around the scene sit a dashboard overlay (capacity by zone, inbound and shipped today, pick accuracy, orders waiting before the next carrier cut-off, a live event feed and a dock timeline), an Inventory table with filters and a 90-day sparkline per SKU, an Orders queue by status and cut-off, and a Docks gantt for the 10 doors. Every change is kept per visitor in the browser, and one button resets the sandbox.

Inventory intelligence is demo AI, and the interface says so: rules and statistics that run in the browser, not a language model. Each SKU has 90 days of generated demand with weekly seasonality; the forecast uses seasonal indices from the last eight weeks and a damped linear trend, safety stock at z = 1.65 sets the reorder point, and four lists flag low stock (raise a purchase order and a truck docks), overstock, dead stock and slotting swaps with the metres they would save. Every action changes the 3D scene: pallets arrive, move, turn violet for clearance or leave.

Everything is procedural: no 3D models, image textures or external requests. The shell is plain JavaScript bundled with esbuild and paints first; the three.js scene is a separate chunk that loads after it. Pallets are instanced meshes with one patched material for carton seams, dimming, heat glow and the intro’s lights-on reveal; shadows are re-rendered only when stock changes, the pixel ratio drops if frames slow down, and rendering pauses in a hidden tab. Phones get a lighter scene with fewer agents, a KPI strip, bottom sheets and a tab bar.

What exists now is a working concept at tarasovvitalii.com/demos/racklight/: four views, four tools and about 230 interface strings in each of English and Russian. We measured the production build on 28 September 2026: the first script is 10 KB gzipped (after the boot rework), the three.js scene 154 KB gzipped, and the whole build 1.0 MB with no image, model or texture files. In headless Chrome on a MacBook Air with an Apple M5, at 1600 × 1000 on a Retina scale, the scene held 55–60 fps under the default 60 Hz frame clock in six states and rendered 160–200 fps, about 5–6 ms a frame, with the cap removed.

We also ran the demo’s own routing code on the 1,140 orders it generates over 30 days. The optimised route was a median 32% shorter than walking the order sheet line by line, and the order shown in the screens drops from 409 m to 211 m. Against a sheet simply sorted by location code the median gain is about 9%, which is the fairer comparison for a warehouse that already sorts its pick lists.

RACKLIGHT is a concept VITON¹³ built to show range, not a commissioned product. Norvane DC-02, its customers, carriers, suppliers, products and every number are fictional and generated in the browser: there are no users, no clients, no warehouse and no revenue behind it, and it proves nothing about adoption or savings in a real operation. The demo AI is rules and statistics on the visitor’s device, not a trained model, and it has never seen real demand.

Nothing here connects to a real WMS, ERP, barcode scanner or dock system; the layout is hard-coded rather than imported from drawings, and routing ignores congestion, carts and batch picking. The frame rates come from one machine in headless Chrome on one day, and the Lighthouse review below was measured on the live address on 28 September 2026, with a simulated phone rather than real devices. The demo is kept out of search on purpose.

How to check a web studio’s portfolio →