Mini Zillow AI Search Engine

Semantic Search for Real Estate Listings using FAISS + Sentence Transformers

This repository contains the reference implementation for the Substack article:
What Is FAISS? Building Your Own Zillow AI Search Engine

A practical, production-minded example showing how to build a mini Zillow-style semantic search engine using modern vector search.

Traditional search engines rely on keywords.
Real users don’t.

This repo demonstrates how to replace keyword search with meaning-based retrieval:

• Turn listings into embeddings
• Index them with FAISS
• Search with natural language
• Retrieve listings based on semantic similarity

No toy demos.
No buzzwords.
Just the engineering.

Client → FastAPI → Embedding Model → FAISS Index → Ranked Listings

Components:

• FastAPI for serving search requests
• Sentence Transformers for embedding listings
• FAISS IndexFlatIP for cosine-style similarity
• CSV listing dataset
• Clean structure ready for MLS integration

• Natural language search
• Vector embeddings (normalized)
• FAISS similarity search (IndexFlatIP)
• Easy to extend with filters
• Clean project layout for real pipelines
• Ready for IVFFlat or HNSW upgrades

git clone https://github.com/<your-username>/mini-zillow-faiss
cd mini-zillow-faiss

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

uvicorn app.api:app --reload

GET http://localhost:8000/search?q=modern+studio+in+new+york

Response:

[
  {
    "score": 0.87,
    "id": 5,
    "title": "Modern studio",
    "city": "New York",
    "price": 550000
  }
]

Use real MLS feeds (BrightMLS, CREA, RESO WebAPI, RETS).

Price, bedrooms, city, property type.

• IndexFlatIP simple and effective
• IVFFlat for large datasets
• HNSW for lower memory and fast recall

Docker-ize and ship to GPU or CPU servers.

Full article with deeper explanations: 👉 What Is FAISS? Building Your Own Zillow AI Search Engine

Covers:

• How FAISS works internally
• Cosine similarity intuition
• Semantic search vs keyword search
• Real engineering pitfalls
• Lessons from production vector search

Built by Alireza Rahmani Khalili
Principal Software Engineer • AI Engineering • DDD • MLS Systems

• Substack: https://nidly.substack.com

• LinkedIn: https://www.linkedin.com/in/rahmanikhalili/

• Personal Website: https://alireza.engineer

Alireza writes about:

• AI engineering

• Vector databases

• Domain-Driven Design

• Real-estate listing systems

• Building nation-scale data pipelines

If you're into engineering that actually works in production, subscribe.

MIT License.