We present the third version of the MemoriEase lifelog retrieval system. This system is a conversational lifelog retrieval system built on an embedding-based retrieval method. For LSC’25, we enhance our system with a RAG approach to the question-answering task. In addition, we also incorporate two embedding models, CLIP and BLIP2, for the embedding-based retrieval. We improve our relevance feedback for visual similarity search by adjusting the query embedding. We describe the results of our system at the LSC’25 challenge, where it achieved third place overall and second place in the QA task. The enhancements in this version of the system improve our performance in the LSC’25 challenge.