Staff AI Platform Engineer · DevOps Tech Lead · Terragrunt Ambassador
A decade running production infrastructure at scale. Now building the infrastructure AI agents run on.
Barcelona 🇪🇸 · still kubectl get pods-ing at midnight, now with an agent that gets there first
Giving AI agents access to production is easy. Making them trustworthy enough to use is the real engineering.
- A company brain agents can trust. An LLM wiki on the Open Knowledge Format standard: agents read it and maintain it, every fact carries its source, owner, and expiry date, and stale knowledge gets flagged instead of served. ~60 tokens always on, everything else loaded on demand.
- Autonomous agents on live infrastructure. One of them replaced me as the first escalation path for "ask DevOps" questions. Read-only investigation by default, human-in-the-loop actions, guardrails that scale with blast radius, and an explicit "I don't know" when the evidence isn't there.
- AI cost architecture. A 3-layer model (token compression proxies, output steering, design-level minimalism) behind a central model gateway, rolled out across 150 engineers. Token spend down 20-80%.
- Agentic SDLC. Spec-driven development with Claude Code, Cursor Cloud Agents, and GitHub Spec Kit, with review gates and security scans keeping AI-generated code production-ready.
- Promtect. My OSS: a local-first proxy in Rust that masks secrets in AI-tool requests before they leave the machine, then restores them in the reply. No root certificate, no telemetry.
- AK Solutions. My AI-automation studio, where the platform thinking meets real AI delivery for engineering teams.
- DevOps Tech Lead at Tikal. I own cloud architecture, engineering standards, reliability, and DevEx across multiple products, mentor the teams, and lead Tikal's DevOps Tech Radar (tech strategy with leaders at CyberArk, Redis, Bluevine).
- Platform engineering at scale. Internal developer platforms (Port.io, Crossplane, GitOps), golden paths, ephemeral environments in under 5 minutes. Dozens of production Kubernetes clusters across EKS, AKS, and GKE, with the SLOs and on-call to keep them honest.
- AI/ML infrastructure. AWS Bedrock and Bedrock Agents, GPU scheduling on Kubernetes (NVIDIA Run:ai, Karpenter), vector databases (Pinecone, Neo4j), all as Terraform/Terragrunt IaC.
- Roots in the trenches. Years of sysadmin and SRE work before "platform engineering" was a job title. I've carried the pager, not just drawn the diagram.
I write about agentic systems, AI infrastructure, and platform engineering on Medium, and speak on AI cost architecture and agent-trustworthy knowledge. A few recent ones:
- I Was the Escalation Path. Now an Agent Is, and It's Stricter Than I Ever Was
- Your Agent's Knowledge Should Cost Sixty Tokens, Not Sixty Thousand
- Agents Wrote Your Docs. Now Answer These Five Questions
- How I Cut My AI Agents Token Bill: A 3 Layers Model
- Ending Shared Dev Envs Chaos: Ephemeral Environments on Demand


