Realest
A voice agent that phones Toronto listing agents for renters and re-ranks their shortlist while they watch.
Problem
Toronto rental listings are often wrong: already leased, parking extra, pet policy unknown. The only way to know is to phone every listing agent yourself and play voicemail tag.
My role
Ranker and live page on a team of 4. I wrote the ranker that orders every shortlist and built the page the renter watches: the sliding reorder, the rent corrections and the detail sheet. I also deployed it on Vercel.
How it works
While the renter talks to the agent, a page texted to their phone shows the shortlist. The agent writes to one session; the page reads it every 1.2 seconds.
Say what you want
The renter phones in and describes the place. The backend ranks 104 Toronto listings and keeps four.
Call every listing at once
On “call them”, one outbound call per listing goes out in parallel, each opening by saying it is an AI.
Re-rank on the answers
Each transcript becomes a typed outcome. Real rent replaces listed rent, leased units sink, and the cards slide into place.
The product
Where it is now
Built in a day at Agents, Everywhere, an AI Tinkerers hackathon in Toronto. Real calls worked on the day, to teammates playing listing agents; the recorded demo uses scripted ones, and the demo board is still online.
- 104Real Toronto rentals, ranked on every brief
- 4Listing agents called at once, one per card
Next project
