Carlos Mendible

In this post, I’ll walk you through the major refactoring of Azure Quick Review (azqr), where I used GitHub Copilot’s plan mode and agent mode while supervising every change. My role was purely architectural: I defined what needed to change, reviewed every proposal, and guided the AI through the process. TL;DR # I refactored Azure Quick Review (azqr) without writing a single line of code by using GitHub Copilot’s plan mode (to design the architecture) and agent mode (to implement it). The refactor eliminated massive technical debt, 72 scanner packages, 72 command files, hundreds of ARM calls and replaced them with a centralized scanner registry, batched Azure Resource Graph queries, a modular pipeline, dynamic command generation, and unified throttling policy.

As organizations scale their Azure OpenAI workloads, throttling (HTTP 429 errors) becomes a critical operational concern. These errors indicate that your requests exceed the provisioned capacity, leading to degraded user experience, failed completions, and potential revenue loss. This post introduces the Azure Quick Review openai-throttling plugin, which helps you identify throttling patterns, analyze affected deployments, and make data-driven decisions for capacity planning.

Azure availability zones are critical for high availability and disaster recovery. However, zone numbers (1, 2, 3) are logical abstractions—their physical datacenter mappings vary across every subscription. Your zone-redundant deployment might actually share infrastructure with your DR environment because different subscriptions map zones differently. This post explores the Azure Quick Review zone-mapping plugin and why zone mappings matter for high availability, disaster recovery, and cross-subscription architectures.

In this post, I’ll show you how to deploy Azure AI Foundry connected to Bing Grounding using Terraform. This setup enables you to leverage Azure’s powerful AI capabilities and enrich them with Bing’s search data, all managed as code for repeatability and automation. Prerequisites # Before you begin, make sure you have:

Last year I wrote a post about Enabling NAP with Terraform. While the post is still valid, I wanted to write about an scenario that many of you might be facing: Enabling NAP when bringing your own VNET. So let’s learn how to create an AKS cluster and enable Node Autoprovisioning (NAP) with Terraform when bringing your own VNET.

In this post, we will explore how to create custom evaluators to evaluate your Generative AI application locally with the Azure AI Evaluation SDK. The results of the evaluation can be uploaded to your Azure AI Foundry project where you can visualize and track the results. Prerequisites # Before you begin, ensure you have the following: