The blog
Ideas on AI, data & engineering
Technical analysis, no fluff. What I learn building with artificial intelligence, told clearly.
Why get ISO/IEC 27001 certified: the trust you build by protecting data
ISO/IEC 27001 is the international information security standard. It's not a certificate to hang on the wall: it's a management system that forces you to analyze risks, put controls in place and continually improve. What it is, why it opens doors for a company, how it's implemented step by step —and how we're adopting it at Xentris Tech—. First of a three-part series.
I was the guinea pig: what I think of Xentris and automating a country starting with yourself
Xentris published an article saying that before selling automation to anyone, they tested it on their own house first. And the guinea pig was me. This is my side: why I believe a country isn't automated from the top down, but one person at a time, and what I stand behind —and what I hold back— about the company I help build.
The «Windows» that AI agents are still missing
Today's AI agents are like MS-DOS: powerful, but only whoever reads the terminal understands what they do. They're missing their «Windows» —the visual layer that lets you see the work—. Why observability is the next big layer, and the dashboard I built to see it.
One agent isn't enough: how orchestrators split work across multiple AI agents
A single AI agent gets overloaded: its context clutters and it gets worse. The orchestrator is the conductor that splits the work across several agents —each on its own front, each result verified— to solve tasks one thread can't handle well. The real patterns, when to use them and when NOT to, with Bob, Claude Code and AutoSGSST as proof.
Say hello to Bob 2.0: when IBM trades autocomplete for an agentic development partner
IBM unveiled Bob 2.0: not a better autocomplete, but an agentic development partner that plans, codes, tests and modernizes enterprise software end to end —with the governance and cost guardrails an enterprise demands. Here's what the leap means, and why it confirms where agent-driven automation is headed.
From app to agent: the AI that reads documents, sees images and writes your reports
For fifteen years, building software meant putting an app on a smartphone. Now AI agents do the work: they read documents, analyze photos and hand you the report as a PDF. AutoSGSST, by Xentris Tech, brings this to workplace-safety compliance in Colombia.
OpenJarvis: the personal AI that runs on your own machine
Stanford released an open-source framework to build AI agents that live on your device, not in the cloud. Privacy, zero cost per token and agents with memory. Here's the good part.
Quantum computing: IBM lets you touch a real one from your browser
Since 2016 IBM has let anyone run code on a real quantum computer over the cloud. Ten years on, with processors like Nighthawk and a roadmap to 2029, here's what you can do today.
NVIDIA DGX Spark: the first personal AI supercomputer
The same software stack that runs NVIDIA's data centers, in a device the size of a game console. A shift in what developers and researchers can do locally.