[Day 2 & 3] re:Invent 24 - Top Announcements and Thoughts

One of my favourite keynotes of re:Invent is always going be the one from Peter DeSantis where he dives deep into the stack and detail the "how" we power foundational services around "Compute", "Storage" and so forth.

One notable announcement from Day 2 was, EC2 P5en instances, optimized for generative AI. P5en is powered by the most powerful NVIDIA H200 Tensor Core GPUs. P5en instances feature up to 8 H200 GPUs which have 1.7x GPU memory size and 1.5x GPU memory bandwidth than H100 GPUs featured in P5 instances.

We saw a many interesting and impactful announcements from Matt Garman, CEO of AWS on Day 3. Highlighting my top favorites below. For ease of navigation, the announcements are organized into categories.

Databases

1/ Amazon Aurora DSQL - Preview

I have been a big fan of distributed databases and this was a much awaited announcement. Aurora DSQL is a multi region, true active-active relational engine that offers customers the ability to build applications that can read and write from 2 regions with minimal latency and strong consistency. Did you hear, strong consistency? Yes. Thats where DSQL shines. Imagine FinTech companies leveraging DSQL to simplify their multi region active active database requirements.Matt also highlighted the fact that DSQL provides 4x faster reads and writes than Spanner. Game changer!

2/Amazon DynamoDB global tables multi-Region strong consistency - Preview

Yet another cool announcement around multi region and strong consistency. Customers were leveraging DynamoDB Global Table for a long time to build active active applications, but with strong consistency being added, customers can expect Zero RPO for their mission critical apps. This feature ensures reading the latest data version from any region and its ideal for applications with strict consistency needs.

Generative AI

3/Amazon Bedrock Model Distillation - Preview

With Bedrock Model Distillation, customers can create smaller, more cost-effective AI models with use-case specific accuracy. Model distillation is an exciting topic and you can leverage models like the Llama 405B model to distill small models. This feature automates the complex process of model fine-tuning and optimization and generates synthetic data from teacher models to train smaller student models.

4/ Amazon Bedrock Guardrails Automated Reasoning checks - Preview

Hallucinations! The buzz word that you normally like to avoid when dealing with LLMs and running them in production. With this addition to Bedrock Guardrails, Automated Reasoning checks detect hallucinations in LLMs and verify response accuracy. Automated reasoning is an intriguing space and many other AWS services like IAM, S3, Verified Permissions leverages its power. Automated reasoning uses mathematical techniques to verify compliance with expert-created policies.

5/ Amazon Bedrock multi-agent collaboration - GA

Welcome to the world of Agents and Agentic flows. Orchestrating and working with agents is always challenging and this is where Bedrock multi agent collaboration eases customers pain. With this feature, customers can now create and manage AI agents working together to solve complex workflows.The feature enables agents to work in parallel to improve response times and precision and reduces development time for complex multi-agent setups.

<Drum roll please> The next one is a big one! State of the Art Amazon Nova Models.

6/Amazon Nova foundation models in Amazon Bedrock - GA

Andy Jassy announced 6 new 1P Amazon Models categorized under Amazon Nova family. 1/ Nova Micro: Text-only model with lowest latency and cost 2/Nova Lite: Low-cost multimodal model for image, video, and text processing 3/Nova Pro: Highly capable multimodal model with best accuracy-speed-cost balance 4/ Nova Canvas: State-of-the-art image generation model and 5/ Nova Reel: State-of-the-art video generation model. These are available in Bedrock as GA.

Nova micro, lite, pro and premier are understanding models and Nova canvas and reels are content generation models.

The super efficient understanding model, Amazon Nova Premier is going to be made available early next year.

I'm already impressed by the quality of images generated by the new Nova Canvas, text to image model. Here is Las Vegas in cartoon style. (Plain simple prompt without any tuning)

7/ Amazon Q Developer Innovations

There were multiple great announcements around Q Developer, and I'm highlighting my top favorites.

Amazon Q Developer can now automate code reviews - Automatically reviews code and provides immediate feedback. Flags suspicious code patterns and provides potential patches. Assesses deployment risk for code changes. This is a huge improvement and excited to take it for a spin.

Amazon Q Developer can now generate documentation within your source code - This is great to save developer time and effort spending cycles for creating documentation for the code they wrote. This feature can automatically generate documentation within source code, addressing developers' time-consuming documentation challenges. The new /doc chat command generates readme files and data-flow diagram.

Amazon Q Developer can now generate automatic unit tests - Most requested enhancement for Q. Amazon Q Developer can now generate automatic unit tests generation to help developers accelerate feature development and improve code quality.Users can generate unit tests by using the "/test" prompt. The agent uses project-specific knowledge to create comprehensive tests.Requires user consent before adding tests to prevent unintended changes.

Amazon Q Developer now provides transformation capabilities for .NET porting - Preview

Amazon Q Developer can now modernize .NET Framework applications to cross-platform .NET. The feature enables porting Windows .NET applications to Linux-ready cross-platform .NET. The feature can accelerate porting process up to 4 times faster than traditional methods with potential to save up to 40% on licensing costs.

GitLab Duo with Amazon Q - Preview

Stocked about this update thats coming to the end to end SDLC lifecycle!

GitLab Duo with Amazon Q, an AI-powered integration that enhances software development workflows within the GitLab DevSecOps platform. Developers can delegate tasks to Amazon Q agents using quick actions. Enables faster feature development and code generation. Provides AI-assisted code reviews and unit test creation. Supports automated Java codebase upgrades from versions 8/11 to 17 and leverages GitLab's unified data store to provide contextual project insights.

Storage

8/Amazon S3 Tables – Fully managed Apache Iceberg tables optimized for analytics workloads - GA

Very excited about Amazon S3 Tables, a new cloud object store with built-in Apache Iceberg support specifically designed for analytics workloads. This feature offers up to 3x faster query throughput and 10x higher transactions per second. This feature introduces "table buckets" for storing tabular data with table-level permissions and provides advanced analytics capabilities like row-level transactions and schema evolution.Includes policy-driven table maintenance for automated operational tasks.

9/Amazon S3 Metadata - Preview

S3 Metadata, will come really handy for data engineering teams as they scale to support multiple business units and personas. This is a new feature for automatically capturing and managing metadata for S3 objects. The feature automatically captures metadata from objects as they are uploaded to S3 buckets. Supports both system-defined and custom metadata like product SKU, transaction ID, etc.Updates metadata tables within minutes of data changes. Very handy capability.

Data

10/ Next generation of Amazon SageMaker

There were multiple interesting innovations in this space and I will quickly summarize them in here. First up is the Next generation of Amazon SageMaker, a unified platform for data, analytics, and AI that provides an integrated development environment for machine learning and data workflows.

Next gen of Sagemaker includes: Amazon SageMaker Unified Studio - Preview for discovering and working with data across tools SageMaker Lakehouse to reduce data silos and unify data from multiple sources and provides native Iceberg support. Integrates tools from Amazon EMR, AWS Glue, Redshift, Bedrock, and existing SageMaker Studio. Accelerated by Amazon Q Developer generative AI assistant.

For ingesting data seamlessly to Lakehouse, there are zero ETL integrations thats available for popular engines like DynamoDB, Aurora and ISV solutions like Salesforce, ServiceNow etc.

Super excited to see the data ecosystem getting unified.

Thats all for now and now lets go #Build.