Creating Contextualized Code Snippets from Natural Language Intents in VSCode

Anonymous

Keywords: NLP, Code Generation, Code Assistant, Code Models

TL;DR: We introduce a novel AI code assistant where multiple code generation models can be plugged in.

Abstract: During software development, developers often turn to code snippets as a starting point for their individual use and further development. These code snippets can be looked up and explored on popular Q&A websites such as Stack Overflow for solving programming-related problems. The context switch between the IDE and the respective Q&A websites required for this affects the programming flow and degrades the developer's programming productivity. We present an extension for the VSCode IDE to reduce the need for context switches. Multiple code generation models built on GPT-3 or Codex can be plugged in and used to generate and contextualize code snippets based on a natural language intent. A qualitative test and a brief pricing analysis indicate that our open-source extension, especially when combined with Codex, can already be a valuable addition to everyday software development and, furthermore, enables developers to benefit from future improvements in natural language processing.