RAG-Powered Property Search Intelligence for Singapore PropTech

The Problem

A fast-growing Singapore PropTech startup had 500,000+ property listings but no intelligent way to search them. Users typed keywords into a rigid filter UI, missing properties that matched their intent but not their exact phrasing. The sales team spent hours manually shortlisting properties — a process that didn't scale. The founders needed a way to let users describe what they wanted in plain English and get accurate, contextual results instantly.

The Solution

NKKTech built a full RAG (Retrieval-Augmented Generation) pipeline. We chunked and embedded all 500K+ listings using OpenAI embeddings and stored vectors in Pinecone. A FastAPI service handled query processing — converting natural-language queries into hybrid vector + metadata searches. The frontend React component provided a chat-like search interface. A nightly sync pipeline kept embeddings fresh as listings changed.

Tech Stack

Results

Key Takeaways

• ✓RAG with hybrid search (vector + metadata filters) dramatically outperforms keyword-only search for unstructured property queries.
• ✓Nightly embedding refresh cycles kept the index current without impacting query latency.
• ✓A custom metadata schema — bedrooms, price range, district, MRT proximity — allowed precise post-retrieval filtering alongside semantic results.
• ✓The system replaced the manual shortlisting workflow entirely, freeing the sales team for higher-value client interactions.