hypernephelist n. [from Greek nepho-, cloud] someone who goes above the clouds.

Secure Code Execution with LLMs: Building CodeBox-AI with Model Context Protocol (MCP)

In recent projects, I’ve been exploring the potential of large language models (LLMs) not just to generate code snippets but also to execute them safely in isolated environments. This combination creates powerful workflows where AI assistants can iteratively develop, test, and refine solutions while providing visualizations and insights from data analysis. The project was of the “AI augmentation” type, where we take the impressive code generation capabilities of models like GPT-4 and Claude, and extend them with a secure execution environment that lets the AI run its own code, evaluate the results, and refine its approach.


Data preparation at scale on Azure Machine Learning using Dask

In a recent project, we worked with our customer on a sustainability project whose goal was to leverage geospatial and climate data to build a platform to perform machine learning model training and inference. The project was of the “machine learning operationalization” type, where we take a codebase that was written by data scientists and machine learning engineers, in the form of experimental Python notebooks, and we port this code to a platform where the processes of data preparation and model training can be automated using workflows and pipelines, and performed at scale.


Adding a trusted certificate for containerd on Kubernetes using a DaemonSet

The Kubernetes project is currently in the process of migrating its container runtime from Docker to containerd, and is planning to obsolete Docker as a container runtime after version 1.20. In most cases, this should be fairly transparent, but if you click through to the Dockershim Deprecation FAQ, you can see a pretty long list of possible ways this could impact you.


Scalable Azure DevOps Agent Pools using VM Scale Sets

When using Azure DevOps Pipelines, you can get started very quickly by using Microsoft-hosted build agents to run your build and test jobs. However there are some cases where custom (a.k.a. “self-hosted”) build agents might be desirable, the typical case being that you need to install a bunch of tools that are not present in the Microsoft-hosted agents. It is always possible to install custom tools on the fly, and it mostly works when all you need is a couple dependencies, however this method will not be practical when you need a full toolchain installed.


Azure VM Custom Script Extensions with Terraform

Terraform provides support for Azure Virtual Machine Custom Script extensions, that are often used to configure a newly created virtual machine and prepare it so it is ready to perform its role. Typical tasks performed in these custom scripts include installing additional packages, configuring system services, creating users, etc.