Prototype
AI Apartments Aggregator

Find apartments ranked by AI photo quality

An experimental tool that collected real estate listings in one place and used a local AI model to rank apartments by how good their photos look — saving hours of manual browsing. The app shows the listings collected so far.

Open App → Read the story
AI Apartments — St. Petersburg
🔍Search listings…
All1 room2 rooms▼ AI Score
🏠AI ★ 9.2
32 000 ₽
Nevsky, 2 rooms, 54 m²
Score
9.2
🏠AI ★ 8.7
28 500 ₽
Vasileostrovsk, 1 room
Score
8.7
🏠AI ★ 6.1
21 000 ₽
Primorsky, 1 room
Score
6.1
🏠AI ★ 3.4
18 000 ₽
Kolpino, 2 rooms
Score
3.4
Ranked by an Ollama vision model — renovation rating 0–10 from the flat's photos
Under the hood

How the pipeline works

Four automated stages from crawl to ranked results, running locally with no external API dependencies.

1

Crawl listings

The Laraue.Crawling library scrapes real estate sites, handles JS rendering, and stores raw listing data into PostgreSQL.
2

Download photos

For each listing, the available photos are fetched, ready for AI analysis.
3

AI renovation rating

A local Ollama vision model looks at all the photos of a flat together, as one collage, and rates its renovation from 0 to 10.
4

Browse & filter

Results surface in the web UI — sort by AI score, filter by rooms and price. Skip past badly photographed listings instantly.
Tech stack

Built with open-source tools

Everything runs locally — no paid APIs, no cloud dependencies.

Ollama

Locally-hosted LLM runner. The project uses a vision model to evaluate apartment photo quality and return a numeric score without sending data to the cloud.

Learn more →

Laraue.Crawling

The in-house C# crawler library powering the data collection. Supports static HTML and JavaScript-rendered pages via PuppeteerSharp.

View project →

ASP.NET Core + PostgreSQL

Backend API written in C# with ASP.NET Core. Listing data and AI scores are persisted in PostgreSQL, served to the web frontend.

Read the article →
The full story

A detailed technical read, months of building

The article covers the crawler, the vision model integration and the ranking formula in detail.

Open source

The crawler library is public

The Laraue.Crawling library used to build this project is open-source on GitHub — reusable for any web scraping project in .NET.

Laraue.Crawling

A highly customizable C# crawler library with support for static HTML (AngleSharp) and JavaScript-rendered pages (PuppeteerSharp). Schema-based extraction and a base class for scheduled crawler jobs.

View on GitHub
Questions

Frequently Asked Questions

It's an experiment — a personal project built to explore AI-powered apartment search. It works and you can browse the collected listings, but expect rough edges rather than a polished product.

Saint Petersburg, Russia, with listings from Cian and Avito. No new listings are collected now, so the app shows a snapshot.

All the photos of a listing are merged into one collage, and a local Ollama vision model rates the renovation of the whole flat from 0 to 10. Separately, an ideality score combines that rating with the floor and the distance to the metro.

No. Crawling, storage, and AI scoring all run locally — there are no paid APIs or cloud dependencies involved.

An experiment

Browse AI-ranked apartments

Browse the snapshot of Saint Petersburg listings collected so far, sorted by AI photo quality. A prototype — but a working one.

Open App → Read the story