An engineering studio for startups whose platforms need to grow up.
MTLabs builds the foundations that let a growing team ship fast without accumulating a decade of technical debt: infrastructure, delivery, security and developer experience.
Platform problems are cheap to prevent and expensive to fix. The decisions a team makes in its first two years decide how fast it moves in its fifth. MTLabs exists to make those decisions early, and on purpose.
MTLabs is the engineering studio of Marco Tomás, Principal Platform Engineer. Fifteen years spent building the platforms that banks, telcos and marketplaces run on: multi-region Kubernetes, zero-trust networks, infrastructure entirely in code.
MTLabs brings that discipline to companies years away from needing a platform team, and who can’t afford to get the foundations wrong in the meantime. Not an agency and not a body shop: one principal engineer, a small number of engagements a year, and the same hands from the architecture diagram to the on-call rotation.
When founders call
It works until that person is on holiday, or leaves. The first thing MTLabs builds is a path anyone on the team can take to production.
Every new engineer costs a week of someone else’s time. Golden paths turn that into an afternoon.
Identity, network boundaries, secrets, audit trails: built in once and answered forever, instead of assembled per deal.
Usually a design problem, not a discount problem. It gets fixed at the architecture, and then it stays fixed.
How it gets built
Paved paths, so every engineer ships without waiting on you.
Safe-by-default guardrails, so the easy way is the right way.
Observability and AI-assisted ops, so problems surface before customers find them.
What that covers
Built with
Engineering work is easy to describe and hard to judge. These are the differences a founder should be able to see within a quarter.
Any engineer ships to production in their first week, without a senior watching.
Environments, access and a working local setup come from a script, not a Notion page.
Identity-first access, network boundaries and audit trails: the evidence SOC 2 and ISO 27001 questionnaires require, ready before procurement asks.
Costs attributed per environment and per team, designed down instead of discounted.
Observability tuned for signal, with AI-assisted triage. On one multi-region estate, average incident resolution fell from about two hours to about thirty-five minutes.
Everything in code, documented, and boring enough to hand over.
Three fixed shapes, each with explicit scope and a written outcome. Nothing open-ended: the continuing engagement runs in quarters, and every quarter is renewed deliberately or not at all.
Teams who suspect the foundations won’t survive the next year of growth, and want to know before the next hire.
A ranked list of what to fix, in what order, and the cost of leaving each one alone.
Post-seed teams putting real customers on the product, still deploying by hand or by hero.
A paved road your team owns: documented, in your repos, with no dependency on MTLabs.
Series A/B teams that need principal-level platform judgement before they can justify a full-time principal salary. Scoped one quarter at a time.
A platform that grows with the team, and engineers ready to run it without the studio.
Every engagement starts the same way: a 15-minute call to establish whether there is a fit. If there is, a longer working session follows, and a written assessment within two business days setting out what MTLabs would tackle first, what it would leave alone, and what the work would cost. If MTLabs isn’t the right partner, you’ll know on the first call, not the third.
Seven positions the studio keeps returning to. Not slogans, but the reason the architecture ends up looking the way it does.
Platforms should disappear.
The best infrastructure is never noticed. Teams ship, all day, without thinking about the road under them.
Complexity is a liability.
Every abstraction has a carrying cost, and it’s paid at 3am by someone who didn’t write it.
Build for next year, not the next decade.
A twelve-person team doesn’t need a bank’s architecture. It needs the handful of practices that prevent the expensive problems.
Trust is earned through consistency.
One reliable deploy proves nothing. A thousand identical ones become a promise the business can plan around.
Developer experience is a product.
Your engineers are the users. Adoption, not elegance, is how a platform gets judged.
The studio should be replaceable.
Everything in code, in your repos, documented. An engagement that creates dependency has failed.
Engineering exists to reduce cognitive load.
Every paved road, guardrail and sane default buys back attention for the product.
Two products that started inside client work: Opsyron, for operational correlation, and Opsyron Security, for cloud-security decisions. MTLabs is the studio; Opsyron is the product line.
Most cloud-security tools surface thousands of findings. Opsyron Security turns them into the short list of actions that remove the most risk, with the evidence, business impact and remediation plan behind each one. Identity-first, across Azure, AWS and GCP.
An operational platform built in-house. Cloudflare, Azure, AWS, GCP, Datadog, GitHub and Kubernetes are normalised into one event model, then related through a graph of services, environments, deployments and alerts. Cloud resources are discovered rather than declared, so the inventory includes what nobody remembered was running.
Not a blog. Notes are written from the work: arguments worth defending, mistakes worth documenting, positions that changed after contact with production. Five are in draft.
The studio’s thesis, and the arithmetic behind it: what prevention costs against what repair costs.
The short list, and the longer list of things founders get sold before they need them.
Adoption is the only score that counts, and most platforms lose it in the first quarter.
Removing friction is. The difference decides what you automate first.
Every abstraction is a loan. A note on the interest rate.
The Lab is the studio’s non-billable half: where patterns get tested before they reach a client, released when they’re worth releasing, and explained in public.
On-call triage tooling, developed against real post-mortems and measured on a live estate.
ActiveWhere k3s beats managed clusters for teams under twenty engineers.
ActiveWhat to measure when the platform’s users are your own engineers.
OpenGo DaemonSet that maps live topology across multi-cloud Kubernetes estates.
Preparing releaseReusable multi-region Terraform modules, extracted from production work.
Preparing releaseAI-assisted triage for Claude Code, built on real post-mortems.
Preparing releaseWhat early infrastructure decisions cost, compounded.
Talk · 45 minWhat adoption actually costs, and who pays it.
Talk · 30 minDesigning a platform your engineers choose to use.
Workshop · half dayFounder of MTLabs. Principal Platform Engineer, fifteen years in. The roles below are employment history: the estates where the patterns MTLabs brings to startups were proven, at a scale where getting them wrong was expensive.
Education
Fifteen minutes to describe what you have and what keeps breaking. If MTLabs is the right partner, a longer session and a written assessment follow. No deck, no discovery phase, no obligation, and an honest answer about fit.