利用 Amazon Nova 多模态嵌入增强多模态搜索和 RAG.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 3 7 8Amit KalawatJames WuPower multimodal search and RAG applications with Amazon Nova Multi

2、modal Embeddings(he/him)Principal Solutions Architect(he/him)Principal AI/ML Specialist SA 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Agenda Vector embeddings overview Amazon Nova Multimodal Embeddings(MME)Multi-modal search design patterns Success stories&resources2 2025,Ama

3、zon Web Services,Inc.or its affiliates.All rights reserved.90%Image,Audio,Video,Documents,SpeechInternational Data Corporation(IDC)indicates 90%of enterprise data is unstructured and growingData is fuel for AI3 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.What are vector embedd

4、ings?0.743,0.720,-0.325,0.195,0.835,-0.945n-dimensional vectorA numerical representation of dataEmbedding model0.243,0.720,-0.65,0.185,0.535,-0.9450.343,0.020,-0.325,0.195,0.835,-0.645TextI love dogs,they areImageAudioVideo4 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Similar

5、vectors=similar meaning and context Vector search with embeddings5 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Large Language ModelPrompt augmentationResponseEmbeddings modelData sourceVector storeEmbeddings modelEmbeddingUserUser InputContext-0.020.89-0.38-0.530.950.17Semanti

6、c searchDocument chunksSemantic SearchData Ingestion WorkflowRetrieval Augmented Generation(RAG)6Embedding powers generative AI applications 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Beyond Text:what about mu