Virtual Software Engineer Agent for Product Requirement Engineering and Management Using GPT-4.1 and CoT

Capturing accurate and complete user requirements remains a persistent challenge in software engineering, often hampered by ambiguous communication and labor-intensive processes. This paper introduces a Virtual Software Engineer Agent designed to support product requirement engineering and management through conversational interaction. Leveraging the capabilities of a Large Language Model (LLM) like GPT-4.1 enhanced by Chain-of-Thought (CoT) reasoning, the agent engages users in natural dialogue, incrementally deriving structured software requirements from unstructured input. Our system aims to address key limitations of traditional methods, including inefficiency, high cost, and misalignment between user intent and technical output. To evaluate its effectiveness, we compare the agent's generated outputs with expert-written requirement documents from a Bangkok-based software firm, using BLEU, METEOR, and ROUGE-L to measure linguistic and semantic similarity. Results indicate that the proposed solution holds strong potential for automating requirement elicitation while maintaining fidelity to user intent and improving overall scalability in requirement engineering workflows.