ammarinjtk.com
GitHub

How It Works

From raw property listings to AI-powered valuations — a five-stage pipeline that processes 70,000+ properties across Thailand.

1

Objective: Scrape 70,000+ property listings from Thailand's leading real estate portals and extract structured data.

42,530 listings · ~$50 total · ~2s per page
2

Objective: Transform raw listing data into ML-ready features: distance calculations, target encoding, government appraisal prices.

30 features (condo) · $3.58 (Google Maps) · ~50ms per listing
3

Objective: Predict fair market price using gradient-boosted tree models trained on 33,000+ real listings.

9.3% MedAPE (condo) · $0 · ~15ms
4

Objective: Find comparable properties, compute nearby market averages, and query government GIS data for zoning and flood risk.

70,220 searchable · $0 (ES Serverless free tier) · ~500ms
5

Objective: Generate expert-quality property analysis that synthesizes the prediction, SHAP factors, comparables, government data, and visual assessment from photos.

~5s per analysis · ~$0.01 per request · 5-20s

Try it yourself

See the pipeline in action with real property data.