School of Engineering Department of Computer Science and Engineering 120 Using Large Language Models (LLMs) for Software Development Supervisor: CHEUNG Shing Chi / CSE Student: WANG Zijing / COMP Course: UROP 1000, Summer This study explores the feasibility of utilizing Large Language Model (LLM)-driven intelligent agents to automate software tool integration and dataset adaptation tasks. Using the open-source log parsing framework loghub-2.0 as the experimental platform, we systematically designed and iteratively optimized targeted prompts to guide different LLM (GPT-4.1, Claude 4, Claude 3.5) driven Agents (Copilot, Cursor) in adapting native log parsing tools to a standardized evaluation framework. Through three versions of prompt design (V1 Basic, V2 Structured, V3 Generalized) and comprehensive experimental evaluation (recording dialogue turns, error counts, and adaptation outcomes), we revealed the critical role of explicit output specifications in prompts and demonstrated the significant advantages of improved prompts (V2/V3) over the basic prompt (V1) in enhancing adaptation success rate and efficiency. This provides a practical foundation and design insights for building more intelligent automated software integration agents. Using Large Language Models (LLMs) for Software Development Supervisor: CHEUNG Shing Chi / CSE Student: WAT Wing Huen / CPEG Course: UROP 1000, Summer This report aims to evaluate the ability of different agents to understand project structures and which component in README is essential to the Agent. Test 1: uses an unmodified README file. The agent only needs to follow instructions to complete the task. Test 2: removes the explicit description of how to run the project while retaining the necessary commands, which Agent can’t strictly follow README, but still has hints to finish the task. Test 3: removes everything from test2, including all necessary commands, except the project setup instructions. This case is designed to test whether the agent can scan through the code and identify the required commands to complete the task.
RkJQdWJsaXNoZXIy NDk5Njg=