Code Generation Techniques in Compound AI Systems and Applications

As AI applications increasingly rely on complex data infrastructure, efficient and reliable code generation for database systems becomes critical. This project focuses on advancing automated code generation tailored for AI-driven workflows, particularly those involving database interactions. We aim to construct a comprehensive benchmark to evaluate the performance and correctness of code generation systems in the context of database tasks. In addition, we propose a novel approach that leverages program synthesis and large language models to automatically generate executable and context-aware code for common database operations. Our goal is to improve developer productivity, reduce manual coding errors, and enable seamless integration between AI models and data systems.