AutoGraphQL: A Bring-Your-Own-Schema Framework for Adapting Open-Weight LLMs to GraphQL APIs

Abstract

AutoGraphQL is a systematic, schema-driven framework for synthesizing GraphQL query datasets directly from GraphQL SDL schemas. It supports a broad range of GraphQL constructs, including fragments, inline fragments, directives, filters, pagination, aliases, and mutations, enabling the automated creation of diverse, executable datasets for fine-tuning LLMs and adapting them to domain or enterprise-specific schemas.

Download our NL2GraphQL corpus used for fine-tuning and benchmarking in the paper.

Please cite the repo if you think our work is helpful to you.

@inproceedings{battu-etal-2026-autographql,
  title     = {AutoGraphQL: A Bring-Your-Own-Schema Framework for
               Adapting Open-Weight LLMs to GraphQL APIs},
  author    = {Battu, Keya and Mehta, Parth and
               Balagopal, Harikrishnan and Kesarwani, Manish},
  booktitle = {Proceedings of the 2026 Conference on Empirical Methods
               in Natural Language Processing},
  year      = {2026},
  url       = {https://nl2graphql-bench.github.io/}
}