This paper explores the use of retrieval-augmented generation (RAG) to support maritime safety analysis by identifying relevant historical incidents from The Nautical Institute when presented with a previously unseen incident. Utilising vector embeddings, the system retrieves similar historic incidents and subsequently pairs them with their associated lessons learnt. This is then used to construct a few-shot prompt which is fed into a generative AI model to produce lessons learnt for a new incident. The approach enables context-specific insights and provides users with access to comparable and referenceable reports. Initial findings suggest this retrieval approach significantly outperforms the current method on The Nautical Institute website to match similar incidents. Furthermore, the lessons learnt generated from Gemma3-4B-IT rivalled both the human response and that of ChatGPT-4.0 Turbo. Notably, the system presented in this paper operates locally without re-quiring internet connectivity, making it suitable for deployment aboard vessels and potentially anticipating likely lessons to steer mitigating actions ahead of an incident occurring.