Finding our next home

Kits Rental Finder

Every 2–3 bedroom rental in Kitsilano, Fairview and Mount Pleasant on one page, with a model that learns which ones we'd actually like.

7 rental sitesUpdated every 3 daysRuns on our Mac mini

The problem it solves

Seven sites, one list

Listings are scattered, and the same unit is often posted on several sites. The finder collects them all and merges the copies.

819listings collected

471unique homes after merging copies

3 daysbetween automatic updates

Craigslist436Zumper92Rentals.ca83 liv.rent83Zillow50Kijiji44RentFaster31

Sites it can't read, like Apartments.com and Facebook Marketplace, can still be added by hand with + Track a listing.

The page

List and map, side by side

Filter by bedrooms, rent, parking, commute time and neighbourhood
Map pins coloured by parking, with SkyTrain and B-Line layers
Works on any device on the home network, and as a Home Assistant tab

The rental finder page: filters along the top, listing cards on the left and a map of Vancouver with listing pins on the right

Every listing

The answers that matter, up front

A listing card showing rent, size, a Recommended badge, parking, availability and commute times to UBC and Downtown

Real commute times

Door-to-door transit to UBC and Downtown from TransLink's timetables: walk to the stop, the route, and at most one transfer.

Parking, decoded

Included, street only, none or not stated, with the ad's exact wording on hover.

What's changed

NEW since your last visit, price changes tracked, removed listings marked, and links to every site posting it.

Built for two

Separate opinions, shared picks

Each of us has our own favourites, hides and recommendations, tied to our Home Assistant logins. Notes are shared.

♥ Recommended

A listing your own model rates highly: in your top 5% and more likely than not to suit you.

♥♥ For you both

Likely to suit both of us. A favourite counts as a yes, and a hide from either of us rules it out.

★★ You both favourited

Shown when we agree. Anything we both favourite or both hide counts double when the models learn.

The recommender

It learns from what you ★ and hide

What it looks at

Facts: rent, bedrooms, bathrooms, size
Commute: minutes to UBC and Downtown
Details: parking, area, laundry, pets, furnished
Wording: up to 400 common words from the ads
Photos: how dated the rooms look

One model each

A simple model that works from a few dozen examples
Listings you haven't marked count as a faint "probably not"
Retrains weekly, only if your choices changed

What you see

A chance you'd like each listing
♥ Recommended for the best few
Reasons on hover, like "shorter UBC commute"
A Best match sort

Because the model is simple, its weights explain each pick.

It reads the photos

Dated kitchens never make the cut

A vision model running on the Mac sorts up to 12 photos per listing by room, then judges whether the kitchen, bathroom and other rooms look dated or renovated.

A listing with no photos, or a clearly dated kitchen, is never recommended
No photo is sent to an outside service

4,845photos analysed across 463 listings

98listings with a clearly dated kitchen

Checked against our choices

The rule matches 12 of the 57 listings we hid, and none of our 30 favourites.

Effect on the picks

Before photos, 4 of 23 recommendations had a dated kitchen. After, none did.

Is it any good?

Better with every choice we make

Each training is tested on marked listings it hasn't seen. At 0.79, it ranks a listing we liked above one we hid about 4 times in 5.

With this few examples, the score moves around. The more listings we mark, the steadier and better it gets.

Test score after each training

0.5 is a coin flip, 1.0 is perfect

Show as a table

Training	Favourites	Hidden	Test score

Quiet and private

It runs itself, at home

Collects new listings every 3 days, in about 10 minutes
Recommendations update after every collection
Everything runs on the Mac mini; the page is only on our home network
To use it: ★ what you like, Hide what you don't, and try the Recommended filter