Nest Agent

Nest Agent

Inspiration

As an immigrant navigating the UK housing market for the first time, I had no mental model for what good value looked like, which neighbourhoods suited my lifestyle, or how to cut through the noise of hundreds of listings. I wasn't alone, millions of first time buyers this every year, without a knowledgeable friend to guide them.

What it does

Nest is an autonomous property search agent. You describe your ideal home in plain English — location, lifestyle, commute, budget, shops you want to have around, vibes and Nest goes to work.

It searches real listings across the web, structures your requirements against a property ontology, and ranks results based on your priorities, not the default filters that most platforms impose.

You can configure how often it runs, so it keeps watching the market and surfaces new matches as they appear.

How I built it

I built Nest using four core tools:

Prometheux — parses natural language queries into a structured property ontology
Tavily — searches real listing sites (Rightmove, Zoopla) and neighbourhood data
Gemini — drives the agent reasoning loop and synthesises results
ClickHouse — stores, ranks, and retrieves the best-matched listings

The agent outputs a cited shortlist in Markdown, with sources, scores, and reasoning for each recommendation.

Challenges we ran into

The hardest challenge was architectural: figuring out how each tool could contribute its strongest capability, rather than just using them as interchangeable search boxes. Getting the workspace set up correctly in Google AI Studio also took meaningful time - huge thanks to Yael for the support there.

Accomplishments that I'm proud of

We took the time to speak directly with each sponsor to understand how to get the most out of their technology. That investment showed, the final integration feels cohesive rather than bolted together, and each tool is doing genuine work in the pipeline.

What I learned

That the real challenge with multi-tool agents isn't connecting the tools, it's designing the handoffs so that each one amplifies the next. We came away with a much sharper intuition for how to compose autonomous pipelines from specialised components.

What's next for Nest Agent

I want to productise this. The immediate next steps are wiring up a payment layer (x402), hosting the API, and validating whether there's real demand at a price point that covers infrastructure costs and leaves room to grow. The goal is to make Nest the go-to tool for anyone navigating an unfamiliar housing market.