AWS announced general availability of TwelveLabs Marengo Embed 3.0 as an embedding model in Amazon Bedrock Knowledge Bases, enabling natural language search over video, audio, and image assets without custom pipelines. The managed service handles segmentation, frame sampling, transcription, and embedding automatically, removing the need to stitch together transcription services, frame extraction, vector databases, and sync logic.
Condé Nast built an internal video discovery tool that uses Amazon Bedrock multimodal LLMs to search inside video content, replacing manual scrubbing through 140,000+ videos across brands like Vogue and Wired. It lets editorial teams find relevant clips without relying only on titles and descriptions.
Condé Nast built an internal video discovery tool that uses Amazon Bedrock multimodal LLMs to search inside video content, replacing manual scrubbing through 140,000+ videos across brands like Vogue and Wired. It lets editorial teams find relevant clips without relying only on titles and descriptions.
AWS published a technical how-to for building an insurance claims assistant on Amazon Bedrock Knowledge Bases, using agentic retrieval to answer natural-language questions over claim documents with citations. The guide combines S3-indexed claim files, metadata filtering, Guardrails grounding checks, and multi-turn conversation. A 40-question evaluation on synthetic data showed 90.5% retrieval recall and 81.2% citation recall.
AWS published a technical how-to for building an insurance claims assistant on Amazon Bedrock Knowledge Bases, using agentic retrieval to answer natural-language questions over claim documents with citations. The guide combines S3-indexed claim files, metadata filtering, Guardrails grounding checks, and multi-turn conversation. A 40-question evaluation on synthetic data showed 90.5% retrieval recall and 81.2% citation recall.
An AWS Machine Learning blog post guides builders choosing a customer-managed vector store for Amazon Bedrock Knowledge Bases, comparing Amazon OpenSearch Service, Amazon Aurora PostgreSQL with pgvector, and Amazon S3 Vectors across three RAG use cases. It includes original benchmarks on OpenSearch Serverless Classic collections using Amazon's ESCI shopping dataset, measuring NDCG@10, latency, and index size across embedding dimension and data type configurations. The guidance helps practitioners trade off latency, cost, and search quality when building RAG on AWS.
An AWS Machine Learning blog post guides builders choosing a customer-managed vector store for Amazon Bedrock Knowledge Bases, comparing Amazon OpenSearch Service, Amazon Aurora PostgreSQL with pgvector, and Amazon S3 Vectors across three RAG use cases. It includes original benchmarks on OpenSearch Serverless Classic collections using Amazon's ESCI shopping dataset, measuring NDCG@10, latency, and index size across embedding dimension and data type configurations. The guidance helps practitioners trade off latency, cost, and search quality when building RAG on AWS.
A new paper introduces Fusion Embedding, a family of models that extend frozen vision-language embedding backbones to handle audio, enabling a single embedding space for text, image, video, and audio. The first generation (fusion-embedding-1) trains only a 16.4M-parameter connector between a frozen audio tower and the frozen base, achieving competitive audio-text retrieval without updating the base model.
A new paper introduces Fusion Embedding, a family of models that extend frozen vision-language embedding backbones to handle audio, enabling a single embedding space for text, image, video, and audio. The first generation (fusion-embedding-1) trains only a 16.4M-parameter connector between a frozen audio tower and the frozen base, achieving competitive audio-text retrieval without updating the base model.