arXiv paper proposes LLM-based system for end-to-end circuit analysis

A new arXiv preprint examines why large language models remain unreliable on engineering problems, using circuit analysis as a test case. The authors note that such tasks demand both multimodal comprehension and exact numerical reasoning, and they describe enhancements to an LLM-based pipeline aimed at solving circuit problems end to end. The work is a revised submission (v3) and reports on improving system performance for this domain.