Towards the Design and Evaluation of Practical Code Generation Tools

This thesis investigates the design and evaluation of automated code generation tools that assist software engineers in real-world development. While many existing tools achieve strong results in controlled settings, their practical applicability remains limited. We propose two tool designs and two evaluations that emphasize practical considerations for tasks ranging from code completion and test case generation to end-to-end software development. The findings highlight the importance of integrating practical aspects into both the design and evaluation processes, suggesting that future research should prioritize not only the effectiveness but also the practicality of the automated code generation tools.