About me
Cybersecurity technician
If you run a small infrastructure and have nobody dedicated to security, I know where you are: the audit that never happens because installing something opens a three-week conversation, the backups nobody has ever restored, the inventory in a spreadsheet. I have been on that side, which is why I build tools that ask nothing in return for using them.
I came from IT support. I started fixing computers part-time and carried on supporting individuals and small businesses, and that is what gives me the judgement to know where an infrastructure breaks before having to defend it. Cybersecurity was where I wanted to end up ever since I read the hacking magazines my father bought me, so between 2024 and 2026 I finished a master’s and the ASIR higher diploma, earned the eJPTv2 and joined the Cantabria Tech Talent programme at UNIR.
Now I work on the security posture of an industrial company and I am designing its cybersecurity training plan. And since November 2025 I build by directing AI agents, with the rules written as files and checkers that enforce them. That is where Talos comes from, which audits how a server is hardened in under ten seconds, and Argos, the platform it was extracted from.
Career
The support years are at the bottom and the security specialisation starts in 2024, on top of them.
Certifications
Only the ones already earned, each with a link to verify it.
eJPTv2
INE SecurityJunior penetration testing certification, entirely hands-on. The exam is a black-box lab: you have to map the network, enumerate services, exploit web and system vulnerabilities, and move laterally until the targets are compromised. There are no multiple-choice questions.
August 2025Verify
What I work with
- AI agents and automation
- Claude Code and Antigravity: agents with permissions, custom skills and verification
- Skills and plans written as data, so the agent does not improvise
- Checkers that break the build when an agent skips a rule
- Delegating across different models for adversarial review
- Ollama and LM Studio for whatever does not leave the machine
- Automating repetitive audit and documentation work
- Defensive security
- Wazuh: deployment, agents, rules and dashboards
- Alert triage and incident documentation
- Log analysis on Windows and Linux
- Incident response
- Endpoint and information protection policies
- Offensive security
- Methodology: reconnaissance, enumeration, exploitation and post-exploitation
- Nmap
- Wireshark
- Burp Suite
- Metasploit
- Hydra
- Netcat
- CrackMapExec
- GoPhish
- Vulnerability analysis
- Nessus
- OpenVAS
- SonarQube: static analysis
- Trivy: dependencies and images
- Semgrep
- Gitleaks
- IriusRisk: threat modelling
- Patch analysis and version diffing
- Systems and virtualisation
- Linux: Debian, Ubuntu, Kali
- Windows client and server
- System hardening
- Proxmox
- VMware
- Docker
- TrueNAS
- Networking and infrastructure
- TCP/IP, name resolution and SSH
- Firewalls and network segmentation
- Virtual private networks
- Nginx and reverse proxying
- Cloudflare
- Terraform
- GitHub Actions
- Jenkins
- Development
- Go
- TypeScript and JavaScript
- Python
- Bash and PowerShell
- SQL: PostgreSQL, MySQL and SQLite
- React and Astro
- Git
- Regulation and frameworks
- Spanish National Security Framework (ENS)
- ISO 27001
- OWASP Top 10
- Risk analysis and management
- Support on compliance audits
How I work
I direct agents, I do not ask them for code
The rules an agent has to comply with I write as files, and on top of them go checkers that break the build when one is skipped. The design, the judgement and the verification are mine; the volume of mechanical work is theirs. That is how Talos has 147 checks written as data.
What audits only reads
Talos writes nothing on the server it scans: not a registry key, not a config file, not a service restart. It can run in production without asking for a maintenance window.
A control that gets in the way ends up switched off
I would rather have one that is followed ninety per cent of the time than a perfect one somebody bypasses on day two. On CybESphere I withdrew two I had put in myself, for exactly that reason.
The backup is a file
In Argos the complete state fits in one: you copy it and you put it back. A fifteen-step restore procedure never gets tested, and an untested backup is not a backup.
I document what fails
A Wazuh agent that stops reporting because its clock has drifted teaches more than the finished dashboard. And in the Metabase lab I started on the wrong version, which cost me an afternoon.