Scholarlink

This study introduces Adaptive Few-Shot Augmentation to optimize input prompts for large language models (LLMs) in software development. Structured prompts incorporating developer specialization, test cases, response format, and task complexity significantly improve code correctness and execution reliability, as demonstrated using HumanEval and CoNaLa benchmarks assessed via CodeBLEU, pass@1, and pass@5 metrics. The method effectively bridges developer expectations and model-generated outputs, enhancing AI-assisted software engineering. Future research will explore automated feature selection and reinforcement learning for further optimization.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Structuring and Optimization of Personalized Context for Large Language Models in Software Developer’s Support

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Michael Vitko,
  • Sergey Kovalchuk

摘要

This study introduces Adaptive Few-Shot Augmentation to optimize input prompts for large language models (LLMs) in software development. Structured prompts incorporating developer specialization, test cases, response format, and task complexity significantly improve code correctness and execution reliability, as demonstrated using HumanEval and CoNaLa benchmarks assessed via CodeBLEU, pass@1, and pass@5 metrics. The method effectively bridges developer expectations and model-generated outputs, enhancing AI-assisted software engineering. Future research will explore automated feature selection and reinforcement learning for further optimization.