Multi-agent AI-assisted Python development system

This paper investigates a Python-based multi-agents programming assistance system built upon Microsoft’s open-source AutoGen framework. To address hallucination issues commonly found in Large Language Models (LLMs), the system integrates the Qwen2.5 model with LoRA fine-tuning. The architecture comprises three agents— Code Generation Agent, Testing Agent and User Agent. These agents collaborate to simulate human-like software development processes, enabling automated code generation, revision, execution and verification. This design enhances code reliability and development efficiency by reducing error rates and minimizing manual testing. This paper demonstrates the feasibility of multi-agent systems in intelligent programming support.