OptoChat: a large language model with retrieval augmented generation for optics

Large language models (LLMs) show strong performance in general text generation and knowledge-based question answering (QA). However, a substantial performance gap remains in optics, a knowledge-intensive scientific field. General-purpose LLMs, trained and tuned without sufficient optics knowledge, often hallucinate when handling optics-related queries. We present OptoChat, a retrieval-augmented LLM purpose-built for the optical domain. By combining domain-specific retrieval-augmented generation with targeted optimization, OptoChat delivers precise, contextually grounded answers for optics QA. To the best of our knowledge, this is the first work to adapt and optimize LLMs specifically for optics knowledge understanding and QA. Experimental results show that OptoChat achieves state-of-the-art performance on optics-focused QA benchmarks and significantly outperforms recent general-purpose models. The interactive website is available at https://optochatai.com .