TwelveLabs Marengo Embed 3.0 now available in Amazon Bedrock Knowledge Bases

Marengo Embed 3.0 Integration

TwelveLabs Marengo Embed 3.0 has been announced as generally available as an embedding model within Amazon Bedrock Knowledge Bases. This integration provides a new option for users to process and search multimodal data.

Addressing Multimodal Search Challenges

Previously, creating semantic search over video and other media assets required complex pipelines involving transcription, frame extraction, embedding models, and vector databases. This new availability aims to simplify this process.

Teams in various sectors, including media, sports analytics, education, security, and retail, often need to locate specific moments within extensive footage using natural language queries.

Amazon Bedrock Knowledge Bases Capabilities

Amazon Bedrock Knowledge Bases is a fully managed Retrieval Augmented Generation (RAG) service. It handles storage, ingestion, embedding, re-ranking, and retrieval for various file types, including MP4, MOV, JPEG, PNG, and audio tracks.

It supports native connectors for Amazon S3, SharePoint, and Confluence. With Marengo Embed 3.0, the service automatically generates multimodal embeddings, capturing visual, textual, speech, and audio signals into a unified vector representation.

Marengo Embed 3.0 Technical Details

Marengo Embed 3.0 is a multimodal embedding model designed to jointly encode video, audio, images, and text. It converts these diverse data types into a compact, 512-dimensional vector space, facilitating efficient semantic search.

Implementation Example

A walkthrough demonstrates creating a knowledge base powered by Marengo 3.0 using the Amazon Bedrock console. The example involves ingesting a 10-minute clip of the 2022 FIFA World Cup final and performing natural language queries against it. Prerequisites include an active AWS account with Bedrock access, an Amazon S3 bucket, and appropriate IAM permissions.