ReactEval: Building LLM Benchmark for Frontend

James Murdza developed GitWit, an AI-driven online editor designed to facilitate the creation of React applications by automating complex coding tasks, especially for less experienced developers. To evaluate the performance of large language models (LLMs) within GitWit, Murdza is also developing ReactEval, an open-source benchmarking tool specifically for frontend applications. ReactEval addresses the challenge of assessing LLMs' effectiveness in generating code by automating the execution of numerous tests using E2B sandboxes, which simplifies the process of building and testing React applications in a web browser. By running hundreds of tests, ReactEval provides insights into LLMs' reliability and helps improve GitWit's performance, specifically in managing dependencies and code generation. GitWit has evolved from a tool that autonomously generates entire applications to a more interactive editor that allows users to contribute their skills alongside AI, making the development process faster and more iterative. Murdza is open-sourcing ReactEval and collaborating with other companies to enhance the tool's capabilities, while also planning to release demo videos and open-source components to further engage the developer community.