JsonCodeExecutor

JsonCodeExecutor is a simple, intuitive tool that enables you to pose natural language questions to your JSON data.

Setup

1. Environment Configuration:

Before starting, ensure you have an .env file in the root directory of your project with the following content:

OPENAI_API_KEY=your_openai_api_key_here

2. Install the required packages:

pip install -r "requirements.txt"

3. Run the main script:

python3 main.py

How It Works

Its actualy very simple.

1. Describing the JSON

We all know that we can't fit a huge json in the context. So what i have done here is instead of directly interacting with large JSON data, JsonCodeExecutor uses the describe_json function. This function recursively traverses the JSON structure and provides a human-readable description. It identifies dictionaries (JSON objects), their key-value pairs, arrays (JSON lists), their elements, and other data types in the JSON.

Example:

json_data = {
    "name": "John",
    "age": 30,
    "city": "New York",
    "hobbies": ["reading", "traveling"]
}

When passed through describe_json, the output will be:

a JSON object which consists of the following keys and types:
- name: a value of type str
- age: a value of type int
- city: a value of type str
- hobbies: an array of a value of type str (all elements in the array have this type)

2. Generate the Required Code

After generating this description, its attached in the prompt for GPT. GPT then crafts the necessary code to fetch the answer that you, the user, are looking for.

Why JsonCodeExecutor?

While searching for solutions to query JSON using natural language, I came across Langchain's JSON Agent here. I found it cumbersome and unnecessarily complex.

To demonstrate this, I've included an example of how Langchain's JSON Agent works in this repository. In comparison, you'll find JsonCodeExecutor to be a more straightforward solution which you can even debug yourself without having to memorize a new word or anything.