Warehouse Digital Twin
A walkable 3D digital twin of a fulfillment center, wired into the operation's own live warehouse and labor systems. It mirrors the real building, refreshed daily rather than as a one-off snapshot.
Two things on this page. What I delivered on the client engagement, and what I have kept building on the synthetic demo since it ended, with newer models. The demo is fictional and further along than what the client saw: people on the floor walk real pick routes from the order data, and you can walk it yourself, first-person, like a game.
The idea
Dashboards tell you the numbers. They do not tell you where, in the actual building, those numbers are happening.
For a physical business, that spatial context is the whole point. So I turned a fulfillment warehouse into a 3D model you can walk through, then put a few key signals right on the floor.
I modeled it like an apartment complex. Every rack is an address, every item a resident, and you can walk the halls and look around.
The important part is that it is not a drawing. I architected the twin around the real facility and tied it to the operation's warehouse management system, which already tracks the digital location of every product.
So the model became a live replica of the actual business, running on their own data refreshed daily instead of a one-off snapshot.
What I delivered on the engagement
- An interactive 3D model of the floor (racks, zones, dock doors, office), modeled to scale from the real facility. Pan, zoom, or drop in and walk it first-person.
- Tied to their live systems. The warehouse system fed product locations and the labor system fed activity, both through the same data warehouse, refreshed daily, so the floor reflected what was actually happening.
- Light KPIs mapped onto the space. Click a zone or a rack and see the handful of signals that actually matter for that spot, including where product should sit based on sales volume.
- A 2D heatmap so density and activity jump out spatially at a glance.
That version was built with Claude Code on the models available in early 2026. It was in review with the team when the contract ended, so it never went into day-to-day use.
What I have added since (demo only)
The engagement ended, the idea did not. In September 2026 I rebuilt the 3D layer of the synthetic demo with Claude Code on Anthropic's newest model (Claude Fable 5.1). Same data shape underneath, so any of it can be pointed back at a live warehouse system. None of this was delivered to the client.
- A building that looks like a building. Steel racking with cartons in every bin position, pallet reserve racks, pack benches on a running conveyor, dock doors with trailers, columns and roof structure. The roof and walls cut away when you look from outside.
- A live floor. Pickers walk real pick routes built from the order data: each order line resolves to a bin, each bin to a spot in an aisle, and a router walks the corridors between them. Forklifts, a reach truck feeding the pick rows from reserve, and mobile robots in the automation cell keep moving. Follow any picker and their route glows on the floor.
- Walk it like a game. First-person with collision, so you move around racks and benches instead of through them. Works with a joystick on a phone.
- Next up: walk distance per order and per worker from those same routes, and what a re-slot would save. The geometry is there, the numbers come next.
The twin keeps the spatial view clean. The heavy operational reporting lives in a companion KPI dashboard.
The platform behind it
The twin is the spatial layer of a broader warehouse ops-intelligence platform I built on the same data warehouse.
A few of the other pieces
- Daily ops cockpit. An auto-refreshed command center (client profitability, labor performance, live alerts) so a manager sees the state of a thin-margin operation the second they log in.
- Scorecards. Ranked report cards for clients, employees, and labor, with monthly client SLAs and cycle-time views.
- Flow snapshot and scenario modeling. Pick a client and a time window to see space usage and SKU velocity, then model "what if" zone consolidations before committing to them on the real floor.
- Health monitoring. A daily job that runs validation checks across the data (date ranges, filters, clients) and pings the team when a number looks off, so leadership can actually trust what the dashboards say.
Why it matters
The real idea is bigger than one warehouse. Any physical business, a store, a facility, a venue, can be a navigable 3D model with its live signals on it.
It connects the data a company already has to the place that data actually describes, which is exactly what a table of numbers can never show you. It is also genuinely useful for a remote team that cannot just walk the floor.
And because it runs on the same data warehouse that pulls in labor and operational data, it is a foundation you keep building on.
You can start to see where product should live based on sales volume, and layer in trends over time, like how people and work actually move through the building.
How it works
Next.js and react-three-fiber render the model straight from the warehouse layout, so the space stays coherent and to scale. Racking and cartons are instanced, textures are generated in the browser, and the roof and walls cut away when you view from outside, so it runs at full frame rate on a laptop with no downloaded assets.
It reads product locations and key metrics from the same data warehouse that unifies the operation's systems, and it stays read-only and lightweight by design.