My top picks of re:Invent 2023
Here is my favorite list, in no particular order...
Abhishek Gupta
Amazon Employee
Published Dec 2, 2023
re:Invent 2023 has been wrapped up. Before we start preparing for the 2024 edition 😅 let me recap the announcements I am most excited about.
As architects/developers/engineers, we love to get into database scalability discussions and right-sizing our database clusters. I know I do. Specially when it comes to Redis, one of my favorite databases! But, with the recently announced serverless option for ElastiCache, we may not have to do all that.
I know, sad but true 🤔 because you can now set up a new cache in under a minute without worrying about choosing a cache node type, number of shards, number of replicas, or node placements across AZs.
Image not found
Who likes configuring IRSA in EKS? I am not a huge fan! Thank god for Amazon EKS Pod Identity that simplifies this experience! Contrary to IRSA, this method doesn't use OIDC identity providers. Once you enable the feature (as an add-on done once per cluster), all you do is create association between the IAM role and the Kubernetes service account you want to associate with your Pod(s).
I have already taken this for a spin (blog post coming soon) and you should too! Check out this AWS News blog and most importantly, the documentation.
Image not found
RAG (Retrieval Augmented Generation) is a powerful technique, but it takes significant effort to build data pipelines to support continuous ingestion from data sources to your vector databases. Knowledge Bases in Amazon Bedrock delivers a fully managed RAG experience. You can pick Amazon OpenSearch Serverless as the default vector database or choose from other available options.
I took it for a spin and it was so much easier to have get datasets ready for QnA - 1/ Configure the Knowledge Base, 2/ load data in Amazon S3, 3/ sync it, and I was ready to go.
Image not found
Check out this great blog post and refer to the documentation once you're ready to give it a shot.
P.S. More content coming soon, especially around how you can use this from your Go applications and some open-source integrations as well. Stay tuned!
There were a bunch of Vector database engines announced this re:Invent!
Now you can use familiar Redis APIs for machine learning and generative AI use cases such as retrieval-augmented generation, anomaly detection, document retrieval, and real-time recommendations. Amazon MemoryDB for Redis now let's you store, index, and search vectors. At the time of writing, it's in preview but I encourage you to try it out.
Check out the documentation for details including the supported vector search commands.
Image not found
Don't forget to activate vector search while creating a new MemoryDB cluster!
Underneath the hood, vector database capability uses k-NN search in OpenSearch. As I mentioned earlier, OpenSearch Serverless is already one of the vector database options in the Knowledge Base for Amazon Bedrock. Now you can create a specialized vector engine–based collection.
The cool thing to note is that Amazon OpenSearch supported about 20 million vector embeddings during preview. Now (post GA), this limit has been raised to support a billion vector scale 🤯
I also recommend checking out this blog post that goes into vector database capabilities in-depth.
Image not found
Vector search is now built into DocumentDB as well. You can set up, operate and scale databases for ML/AI applications without separate infrastructure. Take a look at the documentation to dive in deeper before creating your first vector collection!
Image not found
Agents in Amazon Bedrock aid with development of GenAI apps by providing a platform to orchestrate multi-step tasks. The instructions we provide is used to create an orchestration plan which is then carried out by invoking APIs and accessing knowledge bases (using RAG).
Amazon Bedrock makes it easy to get started with Agents by providing a working draft of an existing agent. Head over to the AWS console to try it out - select the Test button and enter a sample user request.
To dive in deeper, refer to the documentation, including how to use the Agents API.
Image not found
Ok, you love Amazon DynamoDB and have tons of data in it. How do you make it easily searchable? I'm sure you've found different ways of making that happen. But zero-ETL integration simplifies data synchronization from Amazon DynamoDB to Amazon OpenSearch Service - no need to write custom ETL pipelines!
The integration uses DynamoDB export to Amazon S3 feature to create an initial snapshot to load into OpenSearch. After the snapshot has been loaded, the plugin uses DynamoDB Streams to replicate any further changes in near real time.
Read up on the announcement, check out the blog post and follow along in the documentation when you are ready to take it for a spin.
Image not found
Amazon Q is an interactive generative AI-powered assistant trained on over 17 years’ worth of AWS knowledge and experience building in the cloud and it is being integrated into services such as Amazon CodeCatalyst, Amazon QuickSight and more! It's also available in the AWS console and your IDE.
It's set of capabilities are built to support developers and IT professionals alike - be it helping you optimize EC2 instance type selection, build the next serverless app or upgrade your Java apps!
Image not found
(last but definitely, not the least)
Ok, I may not be actively using Rust as I was a while back, but I still have a soft corner for it. And that's why I am happy to see the GA announcement for AWS SDK for Rust ! To be completely honest, I haven't taken it for a (re-)spin yet, but when I do, I will be using the getting started guide to refresh my memory.
The AWS SDK for Rust contains one crate for each AWS service - you can check them out here.
Image not found
That's it for my short (and completely biased) top picks for this year. I couldn't make it to re:Invent this time (last minute emergency 🤷🏽), but I hope to see you all in 2024.
Happy building!
Any opinions in this post are those of the individual author and may not reflect the opinions of AWS.
Comments
Log in to comment