Generating Cypher with an LLM

This program will generate cypher queries and call them on a database that has been populated with dummy data. The cypher queries are generated using Mistral Large as its large language model.

Set Up

• To install all project dependencies (listed in requirements.txt) run pip install -r requirements.txt.
• Obtain a Mistral API key from here

Neo4j Database Set Up

You will need to use Neo4j Desktop, this can be downloaded here

Once installed follow the steps below:

1. Create new project to host your DBMS.
2. Create a new database within the project.
3. Add the username and password to the .env file.
4. Add the database uri to the .env file. The default value for this is bolt://localhost:7687
5. Run the database in Neo4j desktop.

Populate Dummy Data

In order for the application to be run, it must be loaded with the dummy data in dummy_data.txt.

Follow these steps:

1. Open the database in Neo4j desktop with Neo4j Browser.
2. Copy the contents of dummy_data.txt and paste into Neo4j Browser.
3. Confirm that nodes are created by running MATCH (n) RETURN n.

The graph that is populated should look like:

Run the Program

To run the program, confirm that all set up tasks are completed and then run python generate_cypher.py.

An example question has been added: What did I spend more on Sainsburys or Tesco?.

This should return something similar to: John, you spent more at Tesco, with a total of £1750. In contrast, your total spending at Sainsburys was £45.

Please try out different questions to query the database in different ways. There are more example questions below.

Example Questions

• What is the capital of France?
• What transactions were made today?
• How many transactions were made in January 2024?
• Who did I spend the most money with?
• Calculate the total amount Jane Smith spent in each transaction category.
• What was the latest transaction I made to Sainsburys?
• Find all persons who made transactions with Aldi?
• How much have I spent on rent?