GPU cluster
LOADING
Platform engineering · Europe
We design and ship the pipelines, infrastructure-as-code and automation that let your team deploy daily without drama. And when a piece of the product is missing, we build it.
Instruments from a live engagement: a pipeline that moves, a cluster you can explain, alerts that earn the page, and models that stay inside.
Three fronts, one team: the platform, the AI layer, and the product itself.
GitHub Actions, GitLab CI, Argo. Reproducible builds, canary deploys, one-command rollback.
Terraform and Pulumi with versioned modules, identical environments, zero console drift.
Clusters your team can reason about: GitOps, autoscaling, policy, cost under control.
Metrics, logs and traces with SLOs and alerts that only fire when they should.
We move workloads to AWS, Azure or GCP and automate whatever someone still does by hand.
One entry point for every LLM: keys, quotas, caching, tracing and cost control. When data cannot leave, we self-host the models on your infrastructure.
Services, internal APIs and apps built by the same team that will operate them. No handoff to another vendor.
We do not only operate platforms. Some pieces we turn into product. The first is ARTEX.
A planner decides, workers run real tools, and everything flows back into one map of the target. A pentester that never sleeps — inside your walls.
One binary with PostgreSQL, on your own infrastructure. The target's traffic and the evidence do not leave for a third-party SaaS.
A tool that really attacks is only licensed to listed companies and the public sector, under written authorization for every target.
The engagement is designed to end.
We read your repos, pipelines and cloud bills. You get a map of risks and quick wins.
One real end-to-end path to production before we touch anything else.
We extend the pattern across services and teams with reusable modules.
A pass/fail drill on production, then we leave.
Three situations we are hired for. The diagram is the brief.
One path to production — with canary and rollback — before we touch the rest.
Detection, identity, signed artifacts. A report the board can read.
An AI gateway is not a chat box. It is keys, quotas, traces, and a switch.
Tools we have already run in production — not a wish list.
What we actually change in production — not a newsletter.
A pair of H100s does not host this model. An 8× H200 node is ~€33,900 a month; the API at 300 tok/s is ~€870.
Flexera’s 2026 figure is 29% wasted IaaS and PaaS. On an €8,500 bill that is €2,400 a month.
The loaded first-year cost of a hire in Spain, versus a boutique — including the cases where hiring wins.
Article 21 does not wait for the BOE. Detection, identity, signed artifacts, and a board report.
Send two lines about your stack. Within 24h you get an initial read and a ballpark quote — no hour-long discovery call required.