Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: We made a photo search engine for homes for sale
71 points by travisleestreet 36 days ago | hide | past | favorite | 60 comments
We're a small team of 3 engineers, and wanted to make a better way to look for for properties online. Traditional portals (esp. outside of US) are just a bit rubbish for doing anything other than basic searching.

So we've:

- Created a crawler that reads through estate agents websites to find homes for sale

- Parses that through a series of LLMs and other models to understand each home in depth (e.g. floor type, location, total sqft)

- Parses every photo through an embeddings vector space so that people can search for whatever they want.

Check it out: https://jitty.com

Currently only for home sales (not rentals), and only in the UK.

Some examples:

- "Beautiful church conversions up to £1m" https://jitty.com/for-sale/price-up-to-1000000-gbp/detached-...

- "Floor to ceiling libraries with a ladder" https://jitty.com/for-sale/look-for-floor_to_ceiling_librari...

- "Home in london, under £1m, with big beautiful windows" https://jitty.com/for-sale/london/price-up-to-1000000-gbp/de...

- "Bathtubs with an epic view" https://jitty.com/for-sale/look-for-bathtubs_with_an_epic_vi...




We did this in 2013, got serve with a lot of lawsuit. https://en.wikipedia.org/wiki/NeighborCity

Be careful with the real estate industry


I'm curious, how do businesses like Zillow get around this? Do they have direct agreements with MLS?


Zillow, OpenDoor, Redfin, and Realtor.com all pay for API access to the largest Multiple Listing Systems (except for Realtor because they are the ones that are selling the access).

They then use their own “special” algorithms to give their price estimate based on “relevant” factors by what you are describing.

Your real issue is that the agents putting these homes on the market have a financial stake in it selling, and own copyright to all the pictures and descriptions of their listings. So you have more liability if you are scraping their info.

Another issue is who is the target audience for this? Most serious homebuyers are going to use an agent or a trusted partner for a transaction, and are not excited about compounding layers of hurdles to purchase a home

I am a residential real estate appraiser in the US and real estate technologist, which is why I bring up these points.


Zillow has spent years building relationships with thousands of litigious MLS's. It's impossible to build a lawsuit-proof real-estate app in the US without doing that.


They partner with Listhub and pay for the data feed, and/or get deals with individual MLSs and large brokerages. Hopefully, for folks in the UK it’s not as awful as it is in the states.


It seems like this site is more EU based. Does anybody here know if there is an MLS equivalent in this jurisdiction and/or if similar copyright holders over there are as litigious?


There’s not an MLS equivalent in the UK, property data is quite fragmented.

We’re not looking to display anything an estate agent doesn’t want us to - agents tell us they’re very happy for us to serve as an advertising channel. And we’d immediately action any sort of take down request.


Then this might serve as an interesting opportunity for you to sell leads or direct them towards the info you’re scraping from.

Tho if it works as free advertisement on either side, that’s a win-win.


I was on the real state listing site side long time ago, we were trying to block scrapers from getting our data, at the en we loss, and now we pay them to send us users instead of paying Google ads because it’s cheaper. Fun times


"Floor to ceiling libraries with a ladder"

Does that mean it excludes most of the results from "Floor to ceiling libraries without a ladder"?

You know, if I'm buying a house, I think I can supply my own ladder separately...

Less pedantically, what I'm trying to say is: are you really sure these are the kinds of searches that home buyers are really looking for? "Home in london, under £1m, with big beautiful windows" - I suspect that most London buyers are going to care an awful lot about where in London the house is, a city-wide search isn't going to be useful to most. Maybe your functionality (as presented) won't inspire actual buyers.

Speaking of which, that might be a way to improve it - combine with location & mapping data to figure out nearby transport, services, schools, etc...


For sure, those searches are more about showing what’s possible (and making browsing fun) than what a high-intent buyer would actually type in.

That said, from speaking to a lot of London buyers, people are often more flexible on location than you’d expect. The real criteria tend to look more like: “3-bed house under £700k, 30 min commute from [office], near a park, low crime, good schools” rather than “3-bed in Hackney.” Basically along the lines of the location / mapping data you're suggesting.

We’ve already built travel-time search and plan to layer in more of that other 'services' style data in the next few months.


Pretty awesome that you can search for things that would never normally be tagged. "Moldy pool" actually yields appropriate results.


Not from the UK so sorry if this is off-topic, but what are those "dettached" hosues for around 60k? Do they have their own "land"? or you are buying that space "inside" another house? (so in the end you are at the actual owner's mercy?)


Most houses in the UK are "attached" to another property, either row houses or two houses in one. Detached is just what you would call a regular house in the USA -- it's not touching another building and sits on its own land.


Ohhh much simpler than I thought. Are they usually better than attached? (You don't hear your neighbors, for example?)


Basically, yeah. It's your own little fiefdom so you can remodel the whole house how you want, you can tear up and rebuild the outside if you want, you can just not interact with anyone and become a hermit if you want. Your own singular space.


If you’re scraping the estate agents’ sites rather than striking licensing agreements, I can see how you might justify analysing the images to build a property profile but when you’re reproducing them on your site how do you get round the photographer’s copyright?


- I liked the 'surprise' button because I wasn't sure what to type in the search box. I'm so used to typing a location by default. - I like your examples in this HN post and also the categories below the fold. - I like that I can favorite things without registering an account. I've been experimenting with guest accounts and blank states on my site. - When I search for multiple words from the search tab, acts funny. "big windows" became "big". I tried a refresh and the autocomplete said "big windows", but submitting with enter filtered it to "Bath, UK". Maybe there's some timing issue here with fast submits? - I prefer the screenshots / pitch you have on the app store over the website. I like the idea of browsing and filtering, rather than starting with a specific search. Again, that first point that this is something different than location search and helping me understand what I'm supposed to do.

Beautiful site, well done! I love the idea. I'm based in the US.


We've absolutely got some tuning to do on the autosuggest - suspect what you're seeing is some function of a transatlantic trip.


Does your crawler respect robots.txt? Does it request pages with nice delays so that it doesn't bring down servers? Related: https://news.ycombinator.com/item?id=43422413


We don’t respect robots.txt because estate agents often block listing pages - not necessarily to prevent indexing, but for SEO reasons (Google penalises large sites with transient pages).

That said, we do crawl responsibly i.e. we use reasonable request delays, respect rate limits etc. We want agents to like us, ultimately, and blowing up their servers doesn't help with that. If an agent prefers opt-out, we always honour it.


It is an interesting product but I really do wonder if it is something users are asking for. In the market jitty is going for I would say that discovering what is on any given real estate website is not the big pain point.


For sure big real estate portals functionally work. People buy homes through them, even if the experience is clunky. But because they’re entrenched via massive network effects / SEO, and their actual customers are estate agents, there’s basically no incentive to innovate. That's why the top property portals are such great businesses - Rightmove in the UK has like 70% margins.

I think what you're saying is there's that there’s deeper pain further down the value chain? If so, agreed. We think there’s an opportunity to solve some of that, and that building a better product / search experience at the top of the funnel makes it easier to fix problems later in the journey.


I don't know, maybe it depends on what they do with the images. If they just use the images to recreate MLS-style data—the basic stuff like square footage and number of bathrooms—then no, not useful to me. But there are questions that aren't answered by searching MLS data, like: "are there enough electrical outlets in every room", and "does this kitchen have a useful work triangle?". If I could search for that kind of thing, it would be amazing.


Thankfully, this is within UK because Realtor.com is that monopolistic chokepoint of multiple-listing services (MLS) in North America.

If I were you, I quickly package your website for an acquisition by Zoopla and Rightmove.

Future is bright but they dont want you to succeed and may restort to costly litigation, just to stifle you a bit while they get their similar "act" together.

Quick sale and use that proceeds to move on you becoming the next hot acquisition target.


I like the idea! I typed in "futuristic" which got some wild results. The problem is that some properties will have one crazy photo of a single room (sometimes in the lobby of a block of flats) that is nothing like the rest of the property on offer. One property even showed the escalators of a local Underground station as its primary thumbnail, which were kinda futuristic looking.


We run a process on images to try and work out what's what (inside, outside, building, garden etc). Flat with common areas can be quite challenging. Glad you saw a cool escalator, even if we didn't intend to show that to you


Nice one! I would love if it could use coordinates too. Like, "Pool and outdoor shower near me" or "Flower garden with 1h commute to work <drop google marker>"

I know it's hard to figure out time depending on your transportation but at least that would not give me search results from the other side of the country.


That kind of search is already available in many portals. You can actually find the shape of a polígon with the places you can reach in 1h of conmuté easily, as there is a google maps api for it (distance matrix) and finding the properties in a polygon is a solved problem. (


We don't have the drop coordinates part, but you can ask it '...within 1 hour commute by [transport] of [location]' and then it will create a polygon of all the places that match that.


Can you run all your photos through a focal length/depth of field transform to match the human eye? Real estate agents like to bait and switch and waste people's time with glamorous photos that are made using focals lengths that are very different from actual humans.


That's interesting, can you explain a bit more. How would you go about transforming the focal length or depth


Almost as easy as cropping the center. Doing it well automatically is another kind of problem


Improvement suggestion: Keep the search text in the search field when you show the results. The 'what are you looking for' box gets cleared when you show the results, it would be nicer if the search text was kept so that you could tweak it.


Thanks for the feedback. We're still working out the ideal way to manage the search, lots of trade-offs depending on what route you go. But there's definitely room for improvement.


"Bathroom with an epic view": https://jitty.com/properties/TOFkSADP9QuQAV03XTbg?i=78042905


so epic


It's strange that searches add a filter pill with a cross, but a new search just replaces all the filters. It would be great to iteratively build up filters.


Agreed, we're working on this type of additive search at the moment. Just did a bunch of user testing on it yesterday.


Can someone please build something like this for clothes? Because Google refuses to build a search system where I can select a size, color and brand.


Encore is a new-ish used clothing indexer that might be what you're looking for. I also use Gem which doesn't use AI but indexes multiple vintage/used sites and will notify you when something pops up with your saved searches.


Sounds like steps in the right direction, but not entirely what I'm looking for.

I want AI that can scrape shop websites for attributes that people commonly search for, such as size and color, etc., but also shipping methods, shipping costs, etc. I think this would be trivial for an LLM. For me the scope should be bigger than just used clothes. I prefer new clothes (but I wear them until the end). And the system should be web-wide, not just selected shops.

And then I want a basic filtering system that allows me to quickly find what I need by checking some boxes.

It sounds so simple ...


really cool idea. rightmove is awful for searching for houses. I basically can't filter by features at all. though their map search feature is cool cos i might not want to search in an entire county. though i get your probs mvp-ing. really slick though

also there are no houses for Cheshire on your site. replication steps: type cheshire in the search bar and click the suggestion (Cheshire, UK) = 0 results.


Thanks for the kind comments. Just checked Cheshire and you're right, we're missing a polygon for it (as it's a ceremonial county).


Hi, this looks really cool.

What LLM’s are you using ?

And how is this working “Parses every photo through an embeddings vector space so that people can search for whatever they want”

Thank you


For the image search portion specifically, we use Google’s embedding model. We then use vector search (https://cloud.google.com/alloydb/docs/ai/run-vector-similari...) to calculate the distance between the search phrase and pre-calculated embeddings for each image.

Then there’s a bit of ranking and scoring magic to build a results set.


The first property I got back from the 'Surprise' feature was "Penmynydd, Llanfairpwllgwyngyll".

Welsh is extreme.


so many l's and y's


This is amazing. Beautiful design, great categorisation, polished clean UX. Really great work, congratulations


Thanks, appreciate the kind words


Nice work! Curious about your tech-stack, if you don't mind sharing.

Also what are your plans for monetizing, if any?


Nothing more interesting than rails + alloydb (for the vector search) + hotwire.

First time I've built with Hotwire, I'm not a front end eng by trade, but loved the devex.


Do the terms of use on the websites you scraped allow scraping/copying their data?


doesn't UK have something like redfin or zillow?


Aye. https://www.zoopla.co.uk/


Yeah there are a couple big property portals. Zoopla mentioned above is the number 2 player, Rightmove is far and away #1


Well done guys. Looks pretty cool!


thanks!


very nice, what models are you guys using?


Mainly a mix of Gemini Flash 1.5 and Flash 2.0, depending on the task.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: