Back to challenges
Challenge

Property recommendation via K-Means clustering

In a real estate system, related properties needed to be shown on the detail page. K-Means clustering was implemented in an asynchronous Python/Flask service.

JavaScriptExpress.jsPythonFlaskscikit-learn
Property Clustering Pipeline Express.js API Property created/updated async trigger Async Process (background job) HTTP request Python / Flask Service scikit-learn · K-Means p properties → k clusters (k = related count) Clustering K-Means Pipeline 1. Fetch all properties from Express API 2. Feature engineering: price, location, size, type, amenities 3. Normalize features · Run K-Means (k = related count) 4. Return { propertyId: clusterId } mapping HTTP response: clusters Update cluster field Database property.cluster = N Frontend: GET /properties?cluster=n

Context

In a real estate platform (Express.js backend), property detail pages needed to show related/similar properties to improve user engagement and discovery.

Problem

No automated way to determine which properties were similar. Manual tagging was unmaintainable at scale.

Approach

Implemented an asynchronous ML pipeline to compute property similarity clusters:

  1. Trigger: When a property is created or updated, an async process starts
  2. Processing: The process calls a Python/Flask service, sends the property data
  3. Clustering: The service applies K-Means on all properties and returns cluster assignments
  4. Storage: Each property receives its cluster number, stored in the database

The clustering uses k = number of related properties to display per page. With p total properties, the algorithm groups them into k clusters, so each property gets a cluster number. The frontend then queries properties sharing the same cluster to show as related.

Solution

Architecture:

  • Express.js API: Handles property CRUD, triggers async clustering on create/update
  • Python/Flask Service: Receives property data, runs K-Means clustering, returns cluster assignments
  • Database: Stores cluster number on each property record

Clustering Logic:

  • Features: price, location (lat/lon), size, bedrooms, bathrooms, property type, amenities
  • K-Means with k = desired related properties per page (typically 4-6)
  • Runs on full dataset to ensure consistent cluster assignments
  • Async execution keeps property writes fast

Flow:

  1. Property created/updated via Express API
  2. Async process triggered, sends data to Python service
  3. Service runs K-Means on all properties, returns { propertyId: clusterId }
  4. Express API updates each property’s cluster field
  5. Frontend shows related properties by querying same cluster

Result

Related properties are now automatically determined by similarity without manual curation. The async pipeline keeps property writes responsive while the ML service computes clusters in the background.