🏡 HomeMatch-AI

AI-powered personalized real estate search engine using LLMs and vector embeddings.

🚀 About the Project

HomeMatch is an AI-powered real estate search tool that personalizes property listings based on buyer preferences. By leveraging Large Language Models (LLMs) and vector databases, the application enhances standard real estate listings by tailoring descriptions to match users' unique needs.

🔹 Core Features

• 🔍 Smart Search: Uses semantic search to understand natural language queries for property hunting.
• 📝 Personalized Listing Descriptions: AI-generated descriptions that highlight features important to the buyer.
• 🖼️ Multimodal Search: Searches both text and images to enhance property discovery.
• ⚡ Real-time Recommendations: Generates property listings dynamically based on user inputs.
• 🎨 User-Friendly UI: Interactive web application built with Gradio.

🏗️ Project Structure

HomeMatch-AI/
│── chroma_db
│── images/ 
│── .gitignore 
│── HomeMatch.ipynb
│── data_loader.py
│── search.py
│── ui.py
│── README.md 
│── listings.csv
│── listings_cache.pkl
│── requirements.txt 

🛠️ Installation & Setup

1️⃣ Clone the Repository

git clone https://github.com/nedalaltiti/HomeMatch-AI.git
cd HomeMatch-AI

python3 -m venv env
source env/bin/activate  # On Windows use `env\Scripts\activate`

pip install -r requirements.txt

Option 1: Run in Jupyter Notebook

Open HomeMatch.ipynb and run the cells.

Option 2: Run Gradio UI

python app.py

This will generate a local URL to access the app.

Note: The application falls back to an image named default_image.png when a listing has no associated photo. Ensure images/default_image.png exists or set the DEFAULT_IMAGE environment variable to point to a custom file.

⚙️ Configuration

HomeMatch-AI uses a small Pydantic settings class for configuration. Copy the provided .env.example to .env and adjust any values as needed. These environment variables control paths and search behavior:

Variable	Default	Description
LISTINGS_CSV	listings.csv	Path to the property CSV file
VECTOR_DB_DIR	chroma_db	Directory holding the vector database
CACHE_FILE	listings_cache.pkl	Pickle file used for caching the dataframe
IMAGES_DIR	images	Directory containing listing images
DEFAULT_IMAGE	default_image.png	Fallback image when a listing photo is missing
TOP_K	5	Number of search results returned from the vector DB

📊 How It Works

🏡 1. Collect Buyer Preferences

Users enter their desired budget, number of bedrooms, preferred location, must-have features, and property type.

🔎 2. Vector Search in Database

• The entered preferences are converted into vector embeddings.
• A vector search is performed on the database to find the closest matching properties.

🎯 3. AI-Generated Personalized Listings

• LLMs rewrite the property descriptions to highlight features the buyer cares about.
• Images of the listings are retrieved and displayed along with personalized text.

🎨 Demo UI (Gradio)

The project includes a Gradio-powered UI for an interactive real estate search experience.

🚀 Run the UI

python app.py