Project

Neighbourhood OS

Ranks Dutch neighbourhoods by how few conditions for social connection they offer, then drafts a local outreach plan with an LLM pipeline.

What it does

Scores 722 neighbourhoods in Rotterdam, Amsterdam and The Hague with a Connection Desert Index built from CBS, OpenStreetMap and transit data.
Maps the index by borough and breaks each score into vulnerability, access barriers and social infrastructure.
For a chosen borough, an agent pipeline finds real upcoming events, matches them to resident groups and drafts one invitation per channel.

How it works

Offline scripts pull public data into Postgres with pgvector and compute CDI = 0.40 × vulnerability + 0.35 × access barriers + 0.25 × (100 − social infrastructure), normalised within each city. A Next.js dashboard shows the map, and Generate streams every agent step to the browser over Server-Sent Events. It only uses aggregate neighbourhood data, and nothing is ever sent to residents.

The honest bit

A solo 36-hour hackathon prototype, built at EurHackNL 2026 for BCG's "The Lonely City" challenge. The index is not validated against measured loneliness: the weights are hand-set, and in practice it ranks older, less connected outer areas highest. The impact numbers in the interface are illustrative.

Next.js
React
TypeScript
Tailwind CSS
Postgres + pgvector
Drizzle ORM
Server-Sent Events