Skip to content

Support exporting PyDough metadata graph to JSON format #543

Open

Description

@juankx-bodo

It is possible to load a JSON metadata graph into PyDough. However, once the graph is loaded, it cannot be exported back into a JSON format for LLM use.

The knowledge graph contains two types of information: data required to transform PyDough code into SQL, and context required by LLMs to generate PyDough code. For code generation, not all metadata is relevant to the LLM. Therefore, we need the ability to extract the graph back out of PyDough.

NOTE: The primary requirement is simply to be able to export the JSON metadata graph back from PyDough. Once this export capability exists, any formatting, customization, and filtering can be handled externally outside of the DSL.

Proposed Agent Tools
In addition to a full export, having this JSON foundation will allow us to build useful tools for AI agents, such as:

  1. Collection List Lookup: Get a dictionary of all collections and their descriptions in the following JSON format:
{
  "collection1": "description1",
  "collection2": "description2",
  ...
}

  1. Filtered Collection Lookup: Get a dictionary containing only a specific, filtered subset of collections.

Activity

  1. self-assigned this
    on Oct 12, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions