The same Gemma 4 build, vLLM version and GPU served from a SageMaker endpoint and from a plain EC2 instance, on a T4 and an L4: identical decode and answers, a different call path, and what the managed endpoint's 1.40x buys.

I have used CloudWatch Synthetics canaries on real projects, and honestly, they were good...

Gemma 4's 4-bit builds on SageMaker's smallest GPU, an NVIDIA T4, against the L4: a Turing patch for vLLM, the host image the CUDA 13 container needs, speed, memory, answers and cost per token.

Repacking Gemma 4's QAT weights five ways and serving each on the same SageMaker NVIDIA L4 endpoint: int4 linears, int4 embeddings and lm_head, FP8 and int8, across E2B, E4B, 12B, 26B A4B and 31B.

DynamoDB TTL deletes data asynchronously, so learn how to prevent expired “ghost” records from being used while keeping the fix completely free.

Why Amazon FSx for NetApp ONTAP block performance differed by about 2x on the same specified...

Moving on-premises block storage to AWS? The "divide by 625 MBps" sizing approach comes out...

Amazon S3 Files vs. Amazon FSx for NetApp ONTAP: two designs for using S3 as a file,...

Amazon FSx for NetApp ONTAP's NVMe/TCP procedure assumes RHEL 9.3: why multipathing is...

The first three articles covered what Blast Radius does, how to use it, and how it's designed. This...

India has one of the world's largest and fastest-growing cloud, DevOps, Kubernetes, and cloud-native...

SNS and SQS - what a wonderful pair of services. They are very often used together to send messages...

Result I took the AB1-C01 beta on the day it opened and passed with 743 (700 needed). It's...

Ciao 👋 Make some chai, this is a long one, and it has a plot twist I did not see...

I built a synthetic multi-agent loan crew on Amazon Bedrock and made Traccia hard-block a runaway agent, redact applicant PII across every sub-agent, and export EU AI Act audit evidence. Two of my three governance policies blocked nothing - and not because I misconfigured them. Why is the useful part.

Here is the updated, complete blog post incorporating the AWS Console location and Terraform code...

Introduction This is Part 2 of the series "The Data Foundation for AWS Modernization."...

Introduction What comes "next" for VMware workloads? Hardly a day goes by without the...

Introduction I recently attended AWS Community Day Bengaluru on July 11, 2026. There were...

Un iPad no puede correr un contenedor. Eso no ha cambiado. Lo que cambió para 2026 es que dejó de...

Code reviews are an essential part of our day to day work. Their importance and way of execution have...

“ I have checked the documents of AWS for implemented amazon sqs queues for reliable message sending...

Ciao 👋 Here is a sentence that should not be possible: my workflow executed an action a...

Original Japanese article: Icebergは本当にベンダーロックインを解消したのか考えてみた Introduction I'm Aki, an...

I was at AWS Community Day Bengaluru recently on 11 July 2026 and the energy was simply fantastic....

A short background on SageMaker real-time endpoints, then a measured comparison of Gemma 4 E2B's QAT w4a16 checkpoint against the full-size bf16 release on the same NVIDIA L4 endpoint: decode speed, parallel throughput, answers and cost.

Step by step deployment of Gemma 4 E2B to a SageMaker real-time endpoint on one NVIDIA L4 with the AWS vLLM container, driven by the aws CLI and managed by a Python MCP server from Claude Code or Gemini CLI.

Step by step: timing the Rust and Python Iceberg REST clients on the same operations, on a local catalog and on BigLake and OneLake. Rust is 2x to 4x faster per call on the same machine, the two are even over the internet, and Python takes half a second longer to start on every catalog.

Step by step: the Apache Rust Iceberg REST client against seven Iceberg catalogs. It logs in to five of them, and on every one it logs in to, all 13 endpoints it implements answer — 7 reads on three catalogs and 11 writes on the local one, with no failures. The two AWS catalogs need SigV4, which it cannot send, so Rust reaches them through two other crates whose operation sets differ from its own.

The one command that breaks n8n self-hosting on Amazon Linux 2023, why it happens, and the full paste-ready Docker Compose setup that gets you from a bare EC2 box to your first login.

Dans un article précédent sur le VPC partagé, nous avons mis en place une architecture où le...

What the independent measurements of TypeSafe's Jev found in its first eight days: arXiv preprints, GitHub evaluations and blog benchmarks, each traced to its primary source. Accuracy, calibration, speed, cost, failure modes, the prior art, the open alternatives, and what is still unmeasured.

How to design a multi-region, multi-account AWS network on Cloud WAN, from segments and attachments...

Puedes darle a un agente de IA serverless una memoria durable y basada en significado usando búsqueda...

Plain Gemma 4 26B read by its label probabilities against DiffusionGemma's one-step read, both as community 4-bit (AWQ) builds on one EC2 L4, on 1,200 labelled examples and on the 3,880-record public suite where Jev 1.13.0 has published results. Pre-registered, with accuracy, calibration, calibration after 0 to 150 labels, latency and cost.

Run the JSONata ASL you already have against mocked service responses, and check each failure path before you deploy.

I spent a week adding per-agent cost tracing to a multi-agent AWS Bedrock and Strands crew. It returned a perfect answer, showed 200 OK, and still billed ~1.4x. Here is how to catch that silent waste, read-only and at $0.

Introduction Over the past few days, an AI model called Jev has been generating a lot of...

Before reading this article, I recommend familiarizing yourself with elevated roles and directed...

Deep linking looks simple from the outside. A user taps an HTTPS link. iOS recognises the...

Quand on gère des dizaines de comptes et de rôles AWS, ouvrir la Console est un rituel pénible :...

When you manage dozens of AWS accounts and roles, opening the Console is a tedious ritual: find the...

A local-first CLI that scans an AWS account for unused resources, prices each one from the AWS Price List API, and generates or performs the cleanup. Read-only by default, with every price fetched from the live API and dated in each report.

A cost report gets read twice: once by a person in a terminal, once by an agent through an MCP server. Making the server compute every total instead of handing back rows keeps the two answers the same, and measures the report's own arithmetic against the AWS bill.

KiroGraph hits 1.0.0: what clearing a two-month backlog in a single day taught me about...

In my last article https://dev.to/aws-builders/the-secret-ingredient-is-a2a-3dpp, I talked about...

In July 2026, I attended Liz Rice's session, "Detecting Compromised CI with eBPF and Cilium...

Ciao Amici 👋 Make some chai, because the robot accountant is back, and this time I am the one...

🏃 TL;DR In the first article I built semantic product search on Amazon DynamoDB Vector...

The Problem We run a web application on AWS where users upload files – documents, images,...