Overview
Our client is a US real estate agency. Its customers struggled to find the right property among many listings, and its agents needed one place to sell and rent properties, run advertising campaigns and talk to customers. We built a platform that brings these tasks together and uses machine learning to suggest relevant listings to each user.
Solution
The Alltegrio team built a real estate application for buying, selling and renting property. Agents use it to publish listings and run advertising and marketing campaigns for sales and rentals.
- Interactive Map Integration
An interactive map built on the Google Maps API shows each listing at its location, with photos and a description, so users can explore properties for sale or rent by area. The map is updated as listings change.
- Social Media Integration
Social media APIs let agents cross-post a listing to several platforms with one click, so it reaches more people without extra manual work.
- Advanced Filtering Options
Users narrow the search with filters for property type, region and how recently a listing was updated.
- AI and ML-Driven Recommendations
A recommendation model ranks listings for each user from their saved listings, views, filters and enquiries, together with the features of the listings themselves.
- Virtual Property Viewings
Virtual viewings let users look around a property before deciding whether to book a visit.
Technology Stack
Features
- An interactive map that shows every property for sale or rent at its location, with photos.
- One-click cross-posting of listings to social media through their APIs.
- Drop-down filters by property type, region, update date and other criteria.
- Listing recommendations based on each user's behavior and on listing data, ranked with machine learning.
- Virtual property viewings inside the app.

Business Value
The recommender changes how users meet listings. Instead of paging through every result, a user sees the properties that match what they have saved, viewed and asked about, and new listings that resemble them. Agents can then follow up with interested customers from the same platform.
Listings, campaigns and customer contact sit in one application, so agents no longer move between separate tools to publish a property, promote it and answer questions about it.
Final Outcomes
We delivered a real estate application to manage, advertise and market properties. It combines an interactive map, social media cross-posting, filtering, virtual viewings and machine learning recommendations that help each user find relevant listings sooner.
How We Build Property Recommendations
Recommending property is different from recommending products in a shop. Each listing is unique, it disappears once sold or rented, and a buyer may search for months and buy once. Our recommendation systems for listing platforms are built around those facts:
- Listing features. Price, size, property type and amenities are normalized. Location is encoded as distance to transport, schools and the areas a user has already explored on the map.
- Behaviour signals. Saved listings, repeated views, filter changes and contact requests carry different weights. A message to an agent says more than a quick scroll.
- Cold start. New listings have no history, so they are ranked by similarity to listings that drew interest before. New users get results based on their first filters.
- Freshness rules. Sold or withdrawn listings drop out at once, and a price change triggers re-ranking.
- Product layer. We build the web application and its APIs in TypeScript with Next.js and Node.js, and the ranking models in Python with standard ML tooling.
Recommendations are tested offline on held-out sessions first and then in controlled online experiments, because offline accuracy and real enquiries do not always move together.
What Makes Real Estate Platforms Hard
Most buyers now start online. In the National Association of Realtors' 2024 Profile of Home Buyers and Sellers, 43% of buyers said their first step was looking for properties on the internet, and 51% found the home they bought through an online search. A platform carries much of the search, while agents still close most deals, so it has to work for both.
That creates specific problems:
- Listing quality. Duplicates from several agencies, outdated prices and wrong pins on the map damage trust faster than a weak recommendation. Deduplication and validation run before any ranking.
- Sparse data. One user produces few signals before buying, so models rely on listing similarity and location more than on long histories.
- Fair treatment. Housing rules forbid steering users by protected characteristics. Features that can act as proxies for them are reviewed and excluded.
- Map performance. Thousands of markers and frequent updates need clustering and caching, or the map becomes slow on phones.
More on our work in this sector is on the real estate industry and machine learning development pages.
project timeline
core features: live map, one-click cross-posting, AI recommendations
specialists on the team