Real-Estate AI Agent Demo

Streamlit demo for a multi-agent real-estate assistant that searches listings, summarizes market context, and produces concise valuation notes.

What’s included

• Streamlit app UI for collecting user criteria and showing results.
• Direct Firecrawl extraction to collect listing data from selected websites.
• LLM agents to produce market analysis and property valuation summaries.

Architecture / design

1) UI layer (Streamlit)

• app.py renders the form, validates input, and orchestrates the analysis workflow.
• On submit, it calls run_sequential_analysis() and renders results via display_properties_professionally().

2) Data extraction layer

• DirectFirecrawlAgent uses Firecrawl to extract listing data into a Pydantic schema (PropertyListing / PropertyDetails).
• URLs are constructed per selected site (Zillow, Realtor.com, Trulia, Homes.com).

3) LLM agent workflow (sequential)

• Uses the agno framework to build and run agents.
• create_sequential_agents() builds three roles:
  • Property Search Agent (parses Firecrawl results)
  • Market Analysis Agent (brief market insights)
  • Property Valuation Agent (concise per‑property assessment)
• run_sequential_analysis() coordinates the steps and returns structured data used by the UI.

4) API layer (FastAPI)

• api.py exposes GET /health and POST /analyze, wrapping the same run_sequential_analysis() pipeline for programmatic use.

Setup

Prerequisites

• Python 3.10+ recommended
• Firecrawl API key
• OpenAI API key

Create and activate a virtual environment

python -m venv .venv

.\.venv\Scripts\Activate.ps1

.\.venv\Scripts\activate.bat

Install dependencies

pip install -r requirements.txt

Environment variables

You can set the following environment variables in the shell or in a local .env file (recommended). A starter .env file is included.

• FIRECRAWL_API_KEY
• OPENAI_API_KEY
• PORT (optional, REST API only)

Example .env:

FIRECRAWL_API_KEY=your_firecrawl_key
OPENAI_API_KEY=your_openai_key
PORT=8251

Example (PowerShell):

$env:FIRECRAWL_API_KEY="your_firecrawl_key"
$env:OPENAI_API_KEY="your_openai_key"

Run

streamlit run app.py

REST API

This project also exposes a REST API via FastAPI.

Start the API

python api.py

If you prefer uvicorn directly, set the PORT in your shell and pass it through.

Note: To keep the server running in the foreground, start it in a normal terminal session (do not launch it as a detached/background process).

Endpoints

• GET /health → health check
• POST /analyze → run the full analysis

Request shape (POST /analyze)

Fields:

• city (string, required)
• state (string, optional)
• min_price (int, optional)
• max_price (int, optional)
• property_type (string, optional)
• bedrooms (string, optional)
• bathrooms (string, optional)
• min_sqft (int, optional)
• special_features (string, optional)
• selected_websites (array of strings, required)

{
	"city": "San Francisco",
	"state": "CA",
	"min_price": 500000,
	"max_price": 1500000,
	"property_type": "Any",
	"bedrooms": "Any",
	"bathrooms": "Any",
	"min_sqft": 1000,
	"special_features": "Parking, Yard",
	"selected_websites": ["Zillow", "Realtor.com"]
}

Response shape (POST /analyze)

Fields:

• properties (array of objects)
• market_analysis (string)
• property_valuations (string)
• total_properties (int)

{
	"properties": [
		{
			"address": "...",
			"price": "...",
			"bedrooms": "...",
			"bathrooms": "...",
			"square_feet": "...",
			"property_type": "...",
			"description": "...",
			"features": ["..."],
			"images": ["..."],
			"agent_contact": "...",
			"listing_url": "..."
		}
	],
	"market_analysis": "...",
	"property_valuations": "...",
	"total_properties": 2
}

Notes

• The UI allows selecting multiple listing sources and will return any listings extracted.
• Market analysis and valuation outputs are intentionally concise for display clarity.