← All use cases

Halverd

Property search in plain words

Real estateAI integration in a web app

Buyers describe the home they want in one sentence, and AI turns it into filters over the brokerage's listings.

Chapters · Jump to a step

What happens in this video

A buyer types "3-bed villa with a private pool under 6 million" into the search on Halverd's home page. The sentence turns into chips: Buy, Villa, 3+ beds, up to AED 6M, private pool. The results show exact matches first on a list and a map, then near misses that say what they lack. The buyer opens a listing, looks through its photos and facts, and sends a viewing request. On the advisor's side, the request arrives with the buyer's own sentence, and the lead moves along a pipeline from new request to Sold. The video has no sound.

Halverd is a concept brand we designed and built to show this kind of product. The brokerage, its listings and their permits are fictional.

The problem

Property filters are tedious, and buyers describe homes in sentences, not dropdowns. "A three-bed villa with a pool under six million, near a good school" is one thought for a person and eight clicks on most portals.

What we built

  • One-sentence search

    Claude reads the sentence and returns filters, which the server checks against its allowed values before anything is searched.

  • Filters you can see and edit

    Every filter shows as a chip the buyer can edit or remove, with a note where the AI made an assumption.

  • Near misses that explain themselves

    Homes that miss one wish sit below the exact matches and say why: "No private pool, community pool only".

  • Viewing requests

    One short form on each listing, saved together with the buyer's original sentence.

  • An advisor pipeline

    Advisors see what was asked for and move each lead from the first call to Sold.

Under the hood

  • Next.js
  • Claude API
  • Supabase
  • Leaflet

The AI only writes filters — the listings always come from the database, so it can't invent a home, a price or a permit. If the AI is unavailable, a simple keyword-based fallback takes over, and the page discloses this.

Gallery

What it shows

  • We can add AI to an existing kind of product without handing it the data.
  • We design for the day the AI is down: the search still works.
  • We build the business side as well: leads, a pipeline and an admin for the team.

Want this for your business?

Tell us what you have in mind and we'll say honestly how we'd build it.

Start a conversation

Next use case