xbill

A USB tethering diagnostic that costs no cellular data to run, forty measurements across fourteen phones, and one free setting that took the worst run from 15 to 106 Mbps.

Glue and S3 Tables both implement the Apache Iceberg REST catalog specification. One identical request suite against both shows the same totals and thirteen behavioural differences, including two that require opposite things of the same drop request.

One request suite against seven Apache Iceberg REST catalog implementations — Polaris, BigLake, Glue, S3 Tables, Unity, Horizon and OneLake — comparing what each catalog declares it supports against what it actually serves.

chromeos-boot holds two unrelated scripts under one name: stage seeds a real Crostini container from a private bucket, flex skins a bare-metal Debian desktop to look like one. The split exists because Crostini's guest kernel can't load the NVIDIA driver.

vLLM, JAX and PyTorch serving the same Gemma 4 E2B checkpoint on the same AWS G5g GPU, on one harness and one statistic. The decode ranking reverses on boot time. Nineteen instances, four and a half instance-hours, under $3 - which is what let five wrong claims get caught.

vLLM, JAX and PyTorch serving the same Gemma 4 E2B checkpoint on the same AWS G5g GPU, on one harness and one statistic. The decode ranking reverses on boot time. Nineteen instances, four and a half instance-hours, under $3 - which is what let five wrong claims get caught.

A Claude Code skill that turns one markdown file into dev.to, AWS Builder Center, Medium and LinkedIn versions, checks them before they ship, and posts the ones with an API — plus the debugging tools for when a destination mangles something.

A Claude Code skill that turns one markdown file into dev.to, AWS Builder Center, Medium and LinkedIn versions, checks them before they ship, and posts the ones with an API — plus the debugging tools for when a destination mangles something.

Serving Gemma 4 E2B in pure JAX on AWS g5g.2xlarge and g6.2xlarge with a byte-identical payload. The older instance loses 87% of decode to dtype conversion, and nothing in the logs says so.

One hand-written Gemma 4 port, no PyTorch and no vLLM, on two NVIDIA GPUs a generation apart. Most of it ports untouched. Two things do not, and one of them was quietly eating 87% of decode.

Gemma 4 E2B on the two cheapest whole-GPU CUDA instances AWS sells. Same Turing GPU, different host CPU. One is cheaper per hour, the other is cheaper per token, and the reason is not the CPU.

This article is about running a hand-written Gemma 4 port in pure JAX on three different...

This article provides a step by step deployment guide for serving Google's Gemma 4 on an AWS EC2 G5g...

Comparing Agent Cards with A2A - This tutorial aims to fetch the agent card from A2A agents running...

This article provides a step by step look at the A2A agent card a Strands agent publishes on Amazon...

The same research agent built three times — Google ADK on Cloud Run, Strands on Bedrock AgentCore, Agent Framework on Container Apps — all speaking A2A v1.0. Where the platforms differ, where the models differ, and what has to be held constant before either is readable.

A Google ADK agent on Cloud Run, serving A2A to clients that are not ADK — a Strands agent on Bedrock AgentCore and an Agent Framework agent on Container Apps. The card that advertises your bind address, the reply that arrives twice, the event stream once a tool exists, and what Cloud Run brings to the mesh.

A step by step guide to running a Strands agent on Amazon Bedrock AgentCore Runtime, serving the A2A protocol to callers on Google Cloud and Azure, with no AWS access key anywhere.

All three hyperscalers now ship an agent framework, and all three speak A2A. The protocol page will...

A Strands agent on Bedrock AgentCore Runtime serving A2A to callers on Google Cloud and Azure. The AgentCore container contract, the request header it drops, why a session is a cold start, least privilege that actually works, and reaching Bedrock from another cloud with no access key.

One ADK agent on Cloud Run, serving A2A to clients built on Strands and Microsoft Agent Framework, next to two agents that are not Google's. The ADK-specific findings — to_a2a() and the agent card, the reply that arrives twice, the event stream once a tool exists, and what Cloud Run brings to the mesh.

vLLM on TPU cannot load the Gemma 4 E2B QAT exports. So I served them from a hand-rolled JAX engine on a single v6e chip — and found that the thing that actually governs latency is not the model at all.

Step-by-step: scaffolding a real MCP server in Rust with the official rmcp SDK — tools, JSON schemas, AWS calls, stdio transport, testing the protocol by hand, and wiring it into Claude Code.

Step-by-step: scaffolding a real MCP server in Rust with the official rmcp SDK — tools, JSON schemas, AWS calls, stdio transport, testing the protocol by hand, and wiring it into Claude Code.

One FastAPI route, one Converse call with forced tool use, one container service — and the three AWS details that cost me the most time. 20/20 on the fixture set, median 880 ms against the live deployment.

A live-video scanner that barks when it sees a dog, built by a self-paced AI loop over one weekend. Every green checkmark in this project was, at some point, green over something broken.

Step-by-step: getting vLLM's Rust frontend built and running on an aarch64 EC2 G5g box. rustup, setuptools-rust, protoc, the release flag, and how to prove the Rust frontend is actually the one answering.

Step-by-step: getting vLLM's Rust frontend built and running on an aarch64 EC2 G5g box. rustup, setuptools-rust, protoc, the release flag, and how to prove the Rust frontend is actually the one answering.

This tutorial walks through deploying vLLM and some of the key Rust tools used for building and...

Looker hosts its own MCP server now. This walks through connecting Claude Code to it, pairing it with...