Condo GPT is an intelligent assistant for querying and analyzing condominium data in Miami. It uses natural language processing to interpret user questions and provide insights about condo buildings, units, sales, and market trends. Provided is a sample of the Condo Cube database which can be used to query data in the following markets:
- South Beach
- Miami Beach
- South of Fifth
The purpose is to allow real estate agents and investors who are not technical, to easily and quickly perform high-level analyses and comparisons of condos in Miami, and lead their investment decisions. Currently this job is performed by analysts which are paid a salary, so tools like these have the potential to save brokerages thousands of dollars per month in salary.
I used this as a portfolio project to get my first AI Engineer job. I have listed some issues which you can solve or submit your own PR's for improvements to the tool. Open source contributions are a very good way to get AI jobs since technologies like LangChain are so new, and many companies currently do not require work experience with them.
Please feel free to submit a Pull Request.
- Natural language interface for querying condo data
- Integration with Google Maps API for location-based queries
- Dynamic SQL generation for complex database queries
- Interactive maps and charts for data visualization
- PDF report generation capability
- Python 3.x
- Flask web framework
- PostgreSQL database
- LangChain & LangGraph agents
- OpenAI's GPT models for natural language understanding
- Google Maps API for geocoding and directions
- Chart.js for data visualization
- ReportLab for PDF generation
- Clone the repository
- Prepare the environment and database
- virtualenv gpt_env
- Log in to psql as superuser (sudo -u postgres psql on Linux)
CREATE DATABASE condo_gpt;CREATE USER readonly_user WITH PASSWORD 'password';GRANT CONNECT ON DATABASE condo_gpt TO readonly_user;GRANT USAGE ON SCHEMA public TO readonly_user;GRANT SELECT ON ALL TABLES IN SCHEMA public TO readonly_user;ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT SELECT ON TABLES TO readonly_user;\q
- Edit your
gpt_env/bin/activatefile and add the following environment variables
GPLACES_API_KEY: Your Google Places API keyOPENAI_API_KEY: Your OpenAI API keyFLASK_SECRET: Flask Secret Phrase (Can be any string)PG_USER: Your postgres User you set up abovePG_PASSWORD: Password for your postgres userPG_PORT: Port for your postgres database server (default 5432)PG_DB: Name of your postgres database
- Set up the sample data
- Ensure you are in the same directory as sample_db.sql
psql -d condo_gpt < sample_db.sql
- Install dependencies:
source gpt_env/bin/activatepip install -r requirements.txt
- Start the application
python server.py
- Start the Flask server:
python server.py - Open a web browser and navigate to
http://localhost:5000
- Enter natural language questions about Miami condos in the input field
- The system will interpret your question, query the database, and provide relevant answers
- For location-based queries, interactive maps may be generated
- The system can create charts and graphs for data visualization
- PDF reports can be generated for more detailed analysis
- Use the 'Clear Memory' Button to clear the conversation history
- What buildings on collins had the most sales in 2023?
- Generate a graph of this data
- Put the data in an html table
- Add a third column with the total sales volume of each building
- Add a fourth column with the median sales price for each building
- Replace the third column with the closest school to the building, and the fourth column with the driving distance to that school from the building
server.py: Flask application servermain.py: Core logic for processing questions and generating responsestools.py: Custom tools for database querying and API interactionsprefix.py: System message prefix for the AI agentboilerplate.py: Boilerplate code for map and chart generation