César Zubillaga Beraza

Platform & DevOps Engineer

Pamplona / Iruña, Spain · working fully remote · willing to travel · open to EU / US

Six years as a software developer, then six keeping software running. I shipped production application code — PHP and Symfony backends, Angular and React frontends, ecommerce and bespoke internal tooling — before I moved into platform engineering, and I never put the code down, so I debug the service and the cluster underneath it with the same hands: Python, TypeScript and PHP on one side, Kubernetes, GitOps, Terraform and observability on the other. Outside work I run a small production Kubernetes cluster for a trading platform I built end to end — including an autonomous coding agent that picks up issues, opens pull requests and reviews them.

At a glance

6 yrsas a developer — PHP, Symfony, Angular, React, production ecommerce

+ 6 yrsplatform & DevOps on top of it, still writing code daily

End to endmarket data ingest → API → React dashboard → cluster, built solo

Agentsan LLM agent shipping reviewed PRs against a real backlog

Experience

DevOps Engineer · Niji (formerly Neo9 Iberia)

Dec 2022 — present

Madrid-based · now fully remote · 3 yrs 10 mos

Built monitoring and alerting on Prometheus, Grafana and Alertmanager, giving teams real visibility into service health and response times.
Managed infrastructure as code in Terraform with reusable, versioned modules and CI validation on every change.
Automated container image scanning and policy enforcement with Trivy and Kyverno, and integrated secret management — AWS Secrets Manager, Sealed Secrets — into Kubernetes workloads.
Defined deployment strategies (rolling, blue/green, canary) with Helmfile and ArgoCD, and set deployment standards with development teams to shorten delivery times through GitOps and documentation people actually read.
Since 2025, brought LLM agents into that same workflow: drafting and reviewing infrastructure changes against the existing module conventions, so review effort goes to intent rather than boilerplate.
Treat agent output as untrusted input — the same CI validation, policy gates and human review as anything written by hand. The pipeline is the control, not the prompt.

Arcmedia AG

Oct 2016 — Dec 2022

Madrid, Spain · 6 yrs 3 mos

DevOps Engineer

Feb 2020 — Dec 2022

Built CI/CD pipelines on Jenkins and Bitbucket, reaching zero-downtime deployments across Magento 2, Drupal and internal projects.
Automated server configuration and provisioning with Ansible, cutting provisioning time and keeping environments consistent with each other.
Instrumented services and key metrics in Datadog, improving operational stability and bringing mean time to resolution down.

Developer

Oct 2016 — Dec 2022

Kept shipping application code alongside the DevOps work from 2020 on.

Built internal automation tooling end to end, including a catalogue management application for a Swiss jewellery chain in PHP and Angular 2, integrating Redis, MySQL and Elasticsearch over a private network and deployed in Docker containers — keeping product data in sync between internal and external systems.
Developed and maintained client projects on Magento 2 and Drupal: custom modules, data integrations with external systems and the performance work that followed them in production.

Developer · Autocasión (Grupo Vocento)

Sep 2015 — Oct 2016

Migrated the platform to Symfony 2 with MySQL and Redis, improving performance and content management for one of Spain's larger used-car marketplaces.

Independent developer · zbc.space

Nov 2014 — present

Web projects for direct clients: corporate sites, ecommerce, custom content managers and bespoke internal tooling.

Earlier

Zopo Mobile España — software & ecommerce lead, Feb 2014 – Nov 2014. Custom PrestaShop storefront and an order-tracking module built on the carrier's service.
desnaweb — founder, May 2012 – Aug 2014. A technical support business for phone retailers, running on an AngularJS + Symfony SPA I wrote, covering everything from delivery notes to repair tracking.
Flat Comunicación — web developer, Aug 2013 – Feb 2014.

Platform I run outside work

Trading Behaviour Analyzer

A personal system that ingests market data, surfaces destructive trading patterns and validates signals. Six repositories, roughly 218k lines and 300 test files, built and operated by one person. It is the place where I get to own every layer at once — the ingest, the API, the dashboard, the cluster they run on, and the pager when it breaks.

Layer	What it is
Backend	FastAPI + PostgreSQL, Interactive Brokers order flow over `ib_insync`, candles, order book, websocket feed
Frontend	React 19, Vite, Tailwind 4, TanStack Query, Recharts, a three.js viewer — including mobile, debugged on a real handset over USB
Ingest	APScheduler process: screener → LLM or technical signal → Postgres; Alpaca candles; rating enrichment
Platform	k3s, Helmfile, Flux (GitOps, image automation), sealed-secrets, Traefik, Cloudflare Tunnel + Access, self-hosted Gitea with Actions runners
Observability	kube-prometheus-stack, Alertmanager, EFK with Kibana, Jaeger
Agents	An LLM agent that triages issues, posts a plan, implements it and opens a PR; a second agent reviews it. Throttled by a turn-based quota ledger per role.

Incident write-ups

Three failures from that cluster, written up the way I would write them up for a team. They are the honest version: what I thought was broken, and what actually was.

422 restarts from a single character
A = where the kubelet wanted a < put k3s in a restart loop for six hours — while the workloads kept serving traffic, which is exactly what made it hard to see.
The API that its own liveness probe killed
Thousands of "no free DB connection" lines in the log, and every one of them was a symptom rather than the cause. The trigger was a burst from the frontend and a synchronous handler holding a ten-slot pool.
An SSO login page served as a container image
Cloudflare Access answering /v2/ with a redirect broke Flux and Trivy at once. The fix turned on a detail of how the registry derives its auth realm.

Tools

Orchestration: Kubernetes, k3s, Helm, Helmfile, Docker, Traefik, nginx ingress
Infrastructure as code: Terraform (reusable modules, CI validation), Ansible, Kustomize
GitOps & CI/CD: ArgoCD, Flux (incl. image automation), Jenkins, Bitbucket Pipelines, GitHub Actions, Gitea Actions
Observability: Prometheus, Grafana, Alertmanager, Datadog, Elasticsearch/Kibana (EFK), Jaeger
Security: Trivy, Kyverno, Sealed Secrets, AWS Secrets Manager, Cloudflare Tunnel + Access
Languages: Python, TypeScript/JavaScript, PHP, SQL, Bash
Frameworks: FastAPI, Symfony, React, Angular, Magento 2, Drupal
Data stores: PostgreSQL, MySQL, MongoDB, Redis, Elasticsearch
AI in production: LLM agents with quota and audit controls, Anthropic API, agent orchestration and review loops

Education

Telecommunications Engineering · Universidad Pública de Navarra

2009

Specialisation in image and sound: digital signal processing, transmission systems.

Image and Sound · Escuela CTL, Pamplona

2007

Higher vocational diploma, 2,000 hours.

Languages

Spanish: Native
English: C1 — Cambridge C1 Advanced (CAE)
French: B1 — intermediate