Applied AI · business automation · practical training
Applied AI & Automation Consultant — building an independent practice for AI solution discovery, business automation and practical training, backed by long experience in Linux infrastructure, edge systems and production operations.
00 / Position
I got here from the physical layer. Electronics and C first, then a long relationship with GNU/Linux, then eleven years keeping a 24/7 live television signal on air — where the failure is already on screen and the time budget is the length of the incident. That is still the standard I hold a system to: it has to survive being left alone, and someone who did not install it has to be able to operate it.
At AlterMundi — a twelve-person organisation building free infrastructure for community networks — I ran the infrastructure, helped stand up an independent Autonomous System, and architected SAI, a wildfire early-detection pilot whose edge nodes watched the Córdoba hills unattended for months. Across those systems I keep architecture, constraints, review, verification, deployment and operation explicit.
I am now building an independent applied-AI practice focused on researching the right solution before building, automating business processes, and practical AI training. The proposition is new; the engineering record behind it is not.
01 / Evidence
Led technical direction across AlterMundi's production infrastructure and field-deployed systems: brought its infrastructure into an inventoried, documented operating model and established shared architecture, review and verification conventions. Twelve-person collaborative organisation, no direct reports. Cycle concluded August 2026.
Infrastructure-as-code and day-to-day operations for AlterMundi's services on three hosts — a Proxmox VE 9 primary, a legacy PVE/BGP gateway and a public inference host — and about fifteen LXC containers, described in a machine-readable inventory that is the source of truth. Firewall, networking and secrets conventions, backup and restore runbooks, post-mortems, an infrastructure review and a phased hardening plan out of it. Left operable by people who did not install it.
Architect and lead developer of a distributed wildfire early-detection system: edge nodes running computer-vision inference in the field, with fleet management, ingestion and dashboard on top. Three nodes ran continuously for months in Córdoba; field evaluation at the La Rancherita node reached F1 ≈ 0.86 in situ. The detection model was a specialist's work; I built and operated the system around it — edge fleet, ingestion pipeline, inference service and dashboard. My work on the funded pilot concluded when its funding cycle ended. AGPL-3.0.
Tooling and runbooks to operate an independent Autonomous System: announce an IP block to the Internet via eBGP from a border router, join distributed nodes over a ZeroTier mesh, and host public services on community-owned addresses. Main contributor alongside three AlterMundi colleagues.
Citation-first case-building assistant for workers in Iceland: answers only from a versioned authoritative corpus, delegates wage and tax arithmetic to Skattaskyr, and says so explicitly when authority or coverage is missing. Pilot: the corpus covers three collective-agreement families; production certification is pending.
Offline voice dictation for the Linux desktop, built for daily writing rather than demo transcription: a D-Bus daemon owns recording, faster-whisper transcription, SQLite history and clipboard actions; the GTK4 GUI, tray indicator and CLI are thin clients. Built end to end and in daily use as a working tool.
Eleven years keeping a 24/7 live television signal on air — Cablevisión Flow, provincial cable network and streaming platforms — with no maintenance window, resolving incidents while the signal remained live.
02 / Practice
03 / Background
04 / Elsewhere