Parcel

Inspiration

Data centres are being built faster than ever, but the Netherlands is short of grid capacity, space and safe ground. Sites are still often picked on gut feel and a few phone calls. We wanted a tool that lets anyone compare sites on evidence instead.

What it does

Parcel scores any spot in the Netherlands as a data centre site. Drop a pin or search a place, and you get a score built from public data: distance to substations, fibre, water and motorways, and how isolated the site is from towns. Sites in flood zones or near earthquakes get a lower score it came from, and you can change how much eachfactor counts. It also shows plots for sale nearby, and a chat assistant answers questions about the scores.

How we built it

We built Parcel with Next.js, React and TypeScript, Tailwind and shadcn/ui, and Leaflet for the map. The data comes from OpenStreetMap, PDOK, EMSC, Kadaster and IBIS. We a ~500 m grid for the heatmap, and score eachpin live on demand. The chat assistant uses GPT-4o-mini through OpenRouter, and it can only use scores the app has already calculated.

Challenges we ran into

No open land price data: there's no public pertherlands, so we estimated prices from realbusiness-park prices and labelled the listings as synthetic.
Keeping the AI honest: the chatbot sometimes mor the "best site", so we restricted it toranking sites the app had already scored.
Combining very different datasets: we brought water features, plots and earthquake records into one map that stays fast.

Accomplishments that we're proud of

Every score shows its sources and the evidence behind it, not just a number.
You can adjust the weights and see the score c
You can go from "this area looks good" to "here's a plot you could buy" in a few clicks.

What we learned

Trust comes from transparency. Showing where each number comes from, and saying plainly when data is missing, made the tool more convincing than a single score would. We al public geospatial data exists, and how muchwork it takes to combine it.