Low-Code API Generation with LLMs: How GPT-Based Models Are Changing API Development

Abstract

Relying on prebuilt elements and visual interfaces, LCDPs or low-code development platforms are making it far easier and faster to create software. API development is undergoing transformation with the incorporation of large language models (LLMs) such as GPT which makes development is faster, simpler, and more cost effective. Development of APIs requires substantial skill but with LLMs, developers and non-developers can easily create, modify, and integrate APIs with ease.

The study seeks to find out how LLMs assist in the automation processes of testing, documentation, and code generation to create low-code APIs. By incorporating AI, organizations can enhance system integration, reduce human effort and make the workflows more efficient when combined with low-code platforms. This exploration is done with the use GPTEngineer and AutoGen to determine optimal coding techniques through prototyping and testing the feasibility of LLM driven API development.

To automatically generate Java REST APIs, a unique framework is devised known as AutoAPIEval. Instead of the single LLM prompt used for the models chatGPT, stepwise prompting and structured planning are more effective at achieving the desired objective. Performance metrics such as lines of code (LOC) written, along with the ability to create files are used to assess the effectiveness of various LLMs.