Danilo Poccia

The progression goes from raw first principles in pure Python to framework-powered GPU code using PyTorch, JAX, and MLX.

Inspired by Andrej Karpathy's #microGPT, here are eight implementations of the same tiny GPT language model, each version teaching something different about how neural networks are built, trained, and run:

github.com/danilop/micr...

Each pattern involves trade-offs that we discuss so that attendees leave with both practical criteria for their current deployments and a framework for evaluating new approaches as this space continues to evolve.

In this session, we look at possible solutions including deferred loading, progressive disclosure, and runtime controls. These approaches keep detailed definitions out of context until the agent needs them and implement guardrails around how agents can use tools.

A single MCP server can expose tens of tools, and each tool definition consumes hundreds of tokens. Connect a few servers and a significant portion of your context window is gone before any user input is processed. The effect is that AI agent accuracy drops as tool count grows.

Here are the slides for my session today!

Context pressure with MCP: Patterns and trade-offs

danilop.s3.eu-west-1.amazonaws.com/Presentation...

Hey, I’ll be speaking at #MCPconference London on 11–12 February. My session:

Context pressure with MCP: Patterns and trade-offs

There's a free ticket for my network: luma.com/mcplondon202...

Reach out if you're there! I look forward to the tech chat :)

They need help for building more? :)

So building tech debt.

Also, these is an AWS Serverless MCP server that provides info on how to build with a serverless stack. Did you try it?

Very interesting analysis. What I found works best with current agents and models is to give them a link to the reference implementations they should use. So saying something like “use the API Gateway form this sample but with these changes” or “use this Cognito implementation but with this SSO”.

AWS Weekly Roundup: Kiro CLI latest features, AWS European Sovereign Cloud, EC2 X8i instances, and more (January 19, 2026) | Amazon Web Services At the end of 2025 I was happy to take a long break to enjoy the incredible summers that the southern hemisphere provides. I’m back and writing my first post in 2026 which also happens to be my last...

📰🚨 AWS Weekly Roundup: Kiro CLI latest features, AWS European Sovereign Cloud, EC2 X8i instances, and more (January 19, 2026)

#AWSNewsBlog #AWSCommunityDay #reInvent #AWSBuilderCenter #SSACommunity

AWS Weekly Roundup: AWS Lambda for .NET 10, AWS Client VPN quickstart, Best of AWS re:Invent, and more (January 12, 2026) | Amazon Web Services At the beginning of January, I tend to set my top resolutions for the year, a way to focus on what I want to achieve. If AI and cloud computing are on your resolution list, consider creating an AWS…

Here's my AWS Weekly Roundup 👉 AWS Lambda for .NET 10, AWS Client VPN quickstart, Best of AWS re:Invent, and more (January 12, 2026)

ift.tt/iwO2SgI

#AWS #News #Serverless #VPN

How I Used Kiro to Optimize Its Own MCP Configuration I had a problem. My Kiro setup had accumulated 14 MCP servers over time: AWS tools, web automation,...

I had a problem. I had accumulated too many MCP servers over time. So I asked Kiro to improve its own configuration and group tools into Powers that are loaded in context only when needed. Self-optimization is a new pattern.

dev.to/aws/how-i-us...

#AWS #Kiro #DevEx #AI

That’s better!

Hey, Amazon Science is on Bluesky! Welcome :)

@amazon-science.bsky.social

Chain of Draft: Thinking Faster by Writing Less Large Language Models (LLMs) have demonstrated remarkable performance in solving complex reasoning tasks through mechanisms like Chain-of-Thought (CoT) prompting, which emphasizes verbose,…

Here's the Zoom AI Research paper introducing the idea:

Chain of Draft: Thinking Faster by Writing Less buff.ly/UHCYmdK

Move Beyond Chain-of-Thought with Chain-of-Draft on Amazon Bedrock | Amazon Web Services This post explores Chain-of-Draft (CoD), an innovative prompting technique introduced in a Zoom AI Research paper Chain of Draft: Thinking Faster by Writing Less, that revolutionizes how models…

Full implementation details, code samples, and performance metrics in the post:

Move Beyond Chain-of-Thought with Chain-of-Draft on Amazon Bedrock buff.ly/rf6IFOg

CoD works best for structured reasoning where speed and token efficiency matter. For tasks requiring detailed explanations or interpretability, traditional CoT still has its place.

The approach mirrors how humans actually think—brief mental notes rather than verbose explanations. We built a Lambda function to compare standard prompts, CoT, and CoD across foundation models using the Bedrock Converse API.

We tested CoD on Amazon Bedrock with interesting results: up to 76% fewer tokens and 79% lower latency compared to Chain-of-Thought (CoT), with comparable accuracy on logic puzzles and reasoning tasks.

Chain-of-Draft (CoD) prompting is a technique that compresses reasoning steps to 5 words or less while maintaining accuracy.

#AWS #AI

Exploring the zero operator access design of Mantle | Amazon Web Services In this post, we explore how Mantle, Amazon's next-generation inference engine for Amazon Bedrock, implements a zero operator access (ZOA) design that eliminates any technical means for AWS operators…

This represents our continued commitment to raising the bar on security for generative AI workloads handling sensitive data.

Read the full technical details: buff.ly/aT2AUYY

Building on the AWS Nitro System approach, Mantle uses EC2 instance attestation with cryptographically signed measurements from NitroTPM. Customer data stays encrypted through TLS all the way to the Mantle service, where it's processed in a hardened, immutable compute environment.

More details on Mantle, our next-generation inference engine for Amazon Bedrock. Designed with zero operator access (ZOA) from the ground up—no SSH, no Session Manager, no serial consoles. There's no mechanism for any AWS operator to access customer prompts or completions.

#AWS #AI #Security

Plus access to over 30 always-free services with monthly usage limits

After 6 months, you can decide whether to upgrade to a standard AWS account with pay-as-you-go pricing.

A hands-on way to build real cloud and AI skills.

Get started at aws.amazon.com/free

With your Free Tier account, you can explore:
- Compute: Amazon EC2 (VMs), AWS Lambda (serverless functions), Amazon ECS (containers)
- Storage: Amazon S3, Amazon EBS
- Databases: Amazon DynamoDB, Amazon RDS
- AI and Machine Learning: Amazon SageMaker, Amazon Comprehend, Amazon Bedrock

When you create a new AWS account, you receive $100 in credits, with the option to earn up to an additional $100 by completing activities. The Free account plan gives you 6 months to experiment with AWS services—no risk of any charge during this period, and you can upgrade to a paid account anytime.

Happy 2026! If learning AI and cloud computing is on your list of resolutions this year, what about starting with an AWS Free Tier account?

#CloudComputing #AI #AWS #LearnToCode #TechEducation #DevCommunity #2026Goals

Yep, especially considering they have been taken under the London sky!