Platform & infrastructure work

Boring infrastructure
& AI that actually ships.

I'm Mat Strommen, an independent platform and AI engineer. I put AI agents into your dev workflow without the chaos, cut what you're overpaying for cloud, and leave behind infrastructure your team can still read six months from now.

20–30% cloud spend cut 60–70% faster CI +50% team velocity with AI

AI agents · LLMOps · evalsFinOps · AWS · Azurek8s/k3s · Terraform · GitOps

Get in touch AI & software consulting instead?

01 / What this covers

01.1

AI agents in your dev workflow

Claude and GPT agents wired into CI and code review without the YouTube-tutorial energy. Scoped autonomy, regex guardrails before LLM judgment, eval harnesses, and trace coverage so cost and drift don't surprise you. Up to +50% team velocity, no new headcount.

01.2

Cloud cost & AI spend control

FinOps that pays for the rest of the work. Right-sizing, savings plans, ephemeral CI workers, and LLM spend controls baked into the platform so cost discipline is enforced, not hoped for. Typically 20–30% off the monthly cloud bill.

01.3

Platform engineering

k8s/k3s and EKS clusters, Terraform that converges, GitOps delivery, and pipelines that finish in minutes, deterministic guardrails, no allow_failure: true hiding broken tests. The foundation the AI layer runs on. HIPAA/NIST-ready when the vertical needs it.

  1. ME Me I write the brief and I own the merge. Agents open merge requests; nothing reaches production until I approve it, and anything that touches infrastructure always takes a human merge.
  2. AI MODEL Claude crew A supervising Claude Code session plans the work and hands bounded pieces to specialist agents: an architect, builders and testers in isolated worktrees, and a separate auditor that reviews what they wrote.
  3. HARD RULE CI gate Deterministic checks before any model's opinion. Regex catches DROP TABLE, deleted tests fail the build, secrets and high-severity CVEs block, and tests run on every merge request.
  4. HUMAN GATE My merge Green checks are not a merge. The merge request waits for my approval, and infrastructure changes always do.
  5. OUR SOFTWARE Cluster Merged manifests are applied by the pipeline, no clickops. Done means verified where it runs: pods ready and the live URL doing what was asked, not just a green job.
  6. DATA Observe Every agent session is traced. A failed pipeline gets an agent diagnosis, and the fix comes back as a new merge request, never a direct push.
Agents do the work in the open; deterministic gates and my merge decide what ships.

02 / The stack

AI orchestration & agents

  • Claude · GPT · Llama
  • LiteLLM · LangChain
  • Agent SDKs · MCP
  • Eval harnesses
  • Ollama · Bedrock

FinOps & cloud cost

  • AWS · Azure cost modeling
  • Savings plans · right-sizing
  • LLM spend controls
  • Ephemeral CI workers
  • Cost & drift monitors

Platform engineering

  • k8s/k3s · EKS · AKS
  • Terraform · Terragrunt
  • Ansible · GitOps
  • Helm · Docker
  • GitHub Actions · self-hosted CI

Security & compliance

  • HIPAA/HITECH · NIST
  • TLS · secrets management
  • nuclei · trivy
  • Audited agent autonomy
  • Authentik SSO

Observability

  • Prometheus · Grafana · Loki
  • OpenTelemetry
  • GitOps drift detection
  • Langfuse · LLM tracing
  • Alerting that pages on signal

Languages & APIs

  • Python · FastAPI
  • Bash · PowerShell
  • Node.js · React
  • C#/.NET
  • Next.js

Looking for the AI-replaces-SaaS pitch instead? Back to consulting.