AWS re:Invent 2023: Data Analytics Keynote Announcements

Amazon’s Vice President of Database, Analytics and Machine Learning, Swami Sivasubramanian, delivered a fascinating keynote addressing the advancements in GenAI. He focused on the advancements in GenAI over the last year and the symbolic relationship between data, GenAI and humans that’s accelerating transformation for customers. The announcement that resonated with me the most was:

• Amazon Q data integration in AWS Glue is an integration which will significant increase the accessibility of data integration across AWS. It’s a service announcements that I know will have practical use cases for customers.

I’ve put together all the announcements from the keynote and we’re looking to get plenty more over the course of re:invent so keep an eye out for future posts to get across all the announcements.         

GenAI

• Amazon Bedrock support for Anthropic Claude 2.1 which reduces cost of prompts and completions on Bedrock. It has improved system prompts to improve the user experience.
• Model Evaluation on Amazon Bedrock it allows you to evaluate and compare the best foundational model for your use case. It will you to use curated datasets and predefined metrics for automatic evaluations.
• Amazon Bedrock support for Llama 2 70B which supports large scale tasks such as Language Models. It’s fine tuned for chat bots.
• Amazon Titan Multimodel Embeddings this model adapts to unique and proprietary data.
• Amazon Titan Text Lite is model are ideal for summarisation, copywriting and fine tuning.
• Amazon Titan Text Express is a model focused on open-ended text generation, conversational chat and supports RAG.
• Amazon Titan Image Generator which generates studio quality images using natural language prompts. It can be customised to proprietary data. It comes with an invisible watermark to support responsible AI.
• Custom model program for Anthropic Claude is a program will enable you to partner up with a team of experts to assist with customising the powerful models and securely deploy the models to Bedrock in your VPC.

Analytics

• Amazon SageMaker HyperPod is a new distributed capability that can reduce time to train models by up to 40%. It distributes the dataset across thousands of chips in a cluster and process it in parallel. It’s extremely resilient and will automatically replace any faulty compute. Amazon SageMaker Inference reduces model deployment by up to 50%. Amazon SageMaker Studio has a new user experience and capability.
• AWS Clean Rooms ML allows you to apply ML models with your partners without sharing the underlying data. It allows you to deploy the models in a few steps and saves months of development and resources using ready to apply ML models. 

Database & Search

• Vector engine for OpenSearch Serverless is a simple, scalable and high performing vector storage which allows for search to be operated without managing the underlying infrastructure
• Vector search capabilities now available in Amazon DocumentDB , Amazon MemoryDB for Redis and Amazon DynamoDB which provides single-digit millisecond query and update response times. Ten of thousands of queries per second with 99%+ recall
• Amazon Neptune Analytics is an analytics database engine for Amazon Neptune which makes it faster and easier to analyse amounts of graph data. It allows users to find insights from graph data up to 80x faster. Enables faster vector search by storing graph and vector data together.

There’s plenty of announcements to digest that will accelerate drawing insights from your data. The innovations announced by Swami Sivasubramanian are certainly going to accelerate the data analytics adoption of Versent customers as we start to implement these fantastic models and services. If you need to speak to a consultant on how you can utilise these services to transform your business, please do get in contact with us.