project · 2025

Neighbourly, AI location analyser (hackathon project)

Full-stack AI app that takes an address and returns a comprehensive neighbourhood profile (POIs, accessibility, routing to key destinations, market signal, livability score) by orchestrating a multi-step agent workflow over the TomTom MCP servers. Hackathon project that showcases MCP usage.

A full-stack AI app built during an internal hackathon to demonstrate why an MCP-driven agentic stack beats a pile of REST calls when the workflow has more than 3 steps.

What it does

You give it an address. It returns a comprehensive analysis of the neighbourhood:

1. Location profiling: geocodes the address, identifies the surrounding area.
2. POI analysis: discovers and scores nearby points of interest across healthcare, education, shopping, recreation, transport, safety.
3. Accessibility: computes routes and travel times to key destinations with live traffic.
4. Market intelligence: summarises local property pricing trends and investment signals.
5. Livability scoring: composite score across proximity, accessibility, and quality metrics.
6. Visualisation: renders everything on an interactive MapLibre map with dynamic layers.

The whole thing streams. As the agent works, the user sees the map update in real time, partial results appearing, scores firming up.

Stack

• Frontend: React + TypeScript, MapLibre GL for the map, RxJS for stream state, AG-UI for the conversational interface that lets the user interject mid-analysis.
• Backend: Python AG-UI agent built on the Agno framework. Single /agui endpoint that streams agent steps to the frontend.
• Tools: TomTom MCP servers, both tomtom-mcp (search, routing, geocoding) and tomtom-traffic-analytics-mcp (live traffic, junction analytics) consumed as Agno tools.
• Persistence: SQLite for session memory; analyses are resumable across reloads.

Why this earns a spot in projects

The hackathon prompt was “show why MCP matters”. The most compelling answer is not a slide about JSON-RPC, it is an app where the agent autonomously orchestrates 5 tool calls across 2 MCP servers to give a user a coherent answer they could not have got from any individual service. Neighbourly does that. It is also one of the cleanest demos I have built of an AG-UI streaming pattern, where the workflow is the user interface, not just a chat box on the side.

← all projects