WrongMove

WrongMove demonstrates what becomes possible when comprehensive property data is fused with modern AI architectures and real-time 3D visualization. Built entirely on top of Rightmove — the UK’s leading property portal — the platform reimagines how professionals and serious buyers explore the Greater London market.

This is a live, fully functional demonstration rather than a polished marketing site. Every component reflects production engineering decisions made to handle large-scale geospatial datasets, low-latency semantic search, and immersive visualization at city scale.

Core Capabilities

• 01
  Geospatial Intelligence Layer A fully interactive map of Greater London that aggregates live Rightmove listings, price trends, availability, and neighborhood analytics in a single spatial view powered by PostGIS and vector tiles.
• 02
  Advanced Filtering & Query Engine Dynamic slicing of thousands of listings by price bands, property type, size, and precise location — all executed server-side with spatial indexing for sub-second response times.
• 03
  Real-Time 3D Property Visualization Flat listings are transformed into immersive, interactive 3D environments, allowing users to explore interiors and exteriors before shortlisting.
• 04
  Natural Language AI Assistant Powered by a Retrieval-Augmented Generation system that interprets nuanced user intent and surfaces hyper-relevant recommendations across the entire platform.
• 05
  Semantic Vector Search Users describe their ideal home in plain English; Qwen 3.6 embeddings and pgvector deliver genuinely relevant results ranked by semantic similarity rather than keyword matching.

Roadmap

Deeper multimodal AI models for hyper-personalized discovery, predictive pricing, and automated matching. Enhanced neighborhood intelligence with layered 3D context, advanced UI effects, and real-time market sentiment analysis.

End-to-End Technical Architecture

The platform was engineered as a complete modern data and AI stack. OpenStreetMap base layers are processed into optimized mbtiles using Planetiler for high-performance rendering. Rightmove data covering Greater London is systematically ingested via API pagination and stored in PostgreSQL with PostGIS for native spatial operations.

Property descriptions and attributes are embedded using the Qwen 3.6 model within a Python pipeline, creating dense vector representations that are indexed with the pgvector extension. This enables fast, accurate semantic search and powers the RAG system that drives the AI assistant.

On the frontend, MapLibre GL JS delivers buttery-smooth interactive mapping, property markers, dynamic filters, real-time 3D visualization, and fluid animations. Tiles and vector data are served through the Martin tile server, ensuring low-latency delivery directly from the Postgres + mbtiles backend.

Technical Expertise Demonstrated

• Geospatial data pipelines at scale — Planetiler, PostGIS, QGIS, mbtiles, and Martin tile server
• Vector search and RAG systems — Qwen 3.6 embeddings, pgvector, semantic retrieval for property intelligence
• ML models for ranking and personalization — embedding generation, similarity search, and recommendation engines
• Real-time data ingestion and processing — automated Rightmove API scraping, Postgres streaming, and live synchronization
• Full-stack backend and API design — Python services, REST/GraphQL endpoints, and production-grade architecture

If you work in PropTech, geospatial AI, real estate technology, or large-scale data engineering, I would be delighted to connect and exchange perspectives on building the next generation of intelligent property platforms.