Skip to content

Offers & delivery

DevOps, MLOps and Terraform consulting that gets your work through review

For DACH/EU B2B teams. Start with the Production AI Stack Assessment: the straight answer and the evidence inventory in 10 business days. Build with a Production Implementation Sprint or the Private LLM Platform Package, an EU-controlled LLM platform. Add the Platform Operations Retainer only if you need it. No open-ended staff augmentation or six-month strategy decks. No AI workload? DevOps and Terraform work is scoped in the free 30-minute fit call, at the same fixed price per agreed scope.

Book stack assessmentBook a 30-minute fit call
Sample assessment scoresTarget after the 90-day backlog
  1. 1Data & pipelines34
  2. 2Deploy & automation24
  3. 3LLM & agents24
  4. 4FinOps23
  5. 5DevEx34
  6. 6Governance24
Illustrative profile on the 1-5 scale every assessment scores.

Capabilities

Three practices on one foundation

Governed Terraform foundation, DevOps & automation, then MLOps & LLMOps, for AI and non-AI systems alike. The supporting capabilities build on them. Each maps to concrete infrastructure-as-code (IaC), pipeline, and evidence deliverables.

Core service

Infrastructure as reviewed, tested code

Governed Terraform foundation

Terraform and OpenTofu modules built like product code, for any workload on EU regions of AWS, Azure, or GCP, or on a sovereign cloud. We add policy-as-code with Open Policy Agent (OPA), Sentinel, or Checkov, and drift detection. Pipelines stay safe when AI agents open pull requests: plan review, tests, and policy gates before apply. Sovereign-ready landing zones ship with a documented exit. We also migrate Terraform to OpenTofu.

  • Copy-pasted environments and monolithic root modules, with no versioned module registry
  • Local or shared state files without remote backends, locking, or encryption
  • Console changes drifting from code, with no scheduled drift detection
  • Plans applied from laptops: no CI plan review, tests, or policy-as-code (OPA, Sentinel, Checkov)

Core service

Frequent, reversible releases for every workload

DevOps & automation

Release engineering for web platforms, SaaS backends, data platforms, and AI workloads alike. We build CI/CD with progressive delivery and automated rollback, GitOps promotion, and container and Kubernetes platforms. Every build is tested, scanned, and signed in the pipeline, so the security review gets evidence, not a promise. We baseline delivery metrics: deploy frequency, lead time, change-failure rate and time to restore. Canary releases ship with signed artefacts and a software bill of materials (SBOM).

  • Click-ops deploys for application, data, and model changes
  • No GitOps or clear pipeline ownership across services
  • No delivery-metrics baseline, so nobody can show delivery is improving
  • All-or-nothing releases with no canary or blue-green path
  • Container images unscanned and unsigned, with no SBOM for procurement

Every change, the same path

  1. Commitmerged pull request
  2. CI testsunit, evals, policy
  3. Scan, SBOM, signsupply-chain evidence
  4. Canarysmall traffic slice
  5. Productionpromoted by GitOps

Canary fails its health check: automatic rollback, the stable version keeps serving

DORA metrics, baselined at kickoff and tracked per release

  • Deploy frequencytarget: up
  • Lead timetarget: down
  • Change-failure ratetarget: down
  • Time to restoretarget: down

Core service

Models, prompts, and agents that ship like software

MLOps & LLMOps

Training pipelines, a model registry, evaluation gates, and promotion workflows for classic ML. For large language models (LLMs): a gateway with routing, budgets, and per-tenant cost tags, prompt and tool registries, and evals in CI. Every model, prompt, and agent reaches production through the same reviewed path and is monitored once live.

  • No model registry or eval gates before production promotion
  • Training and serving feature skew across environments
  • Manual deploy runbooks with no rollback path

Gateways, evals, and agents under control

LLM gateway & agent controls

An LLM gateway with auth, routing, budgets, and per-tenant cost tags, plus prompt and tool registries. Retrieval-augmented generation (RAG) and agent evals run in CI, and each agent's tool permissions are scoped, including Model Context Protocol (MCP) tools. Inference becomes observable, budgeted, and releasable.

  • Shadow API keys bypassing gateway budgets and logging
  • RAG pipelines without faithfulness evals in CI
  • Agents and MCP tools without auth or approval harnesses
  • Missing cost tags on inference routes and tenants

Gateway, guardrails, logging, and cost control

EU-controlled LLM platform

One platform for every LLM call: a gateway with guardrails, audit logging, and cost control. Routed to EU-hosted APIs, or to self-hosted open-weight models (vLLM or equivalent) where residency or model control demands it. Keys you hold, open components, and one infrastructure-as-code pattern for EU regions, sovereign clouds, or bare metal.

  • SaaS-only LLM dependency with no self-host exit path or residency story
  • GPU hosts provisioned by hand, with no Terraform/Pulumi for cloud or bare metal
  • Fine-tunes and adapters outside a registry with no promotion gates
  • RAG demo without vector-store ops, backup, or CI faithfulness evals

AI controls on your Terraform foundation

AI landing zones

The AI layer on top of your Terraform foundation, inside the EU boundary: isolated accounts for model workloads. GPU quotas and node pools with guardrails, and private networking and workload identity for model endpoints. We add data-access boundaries for training and RAG, and inference logging sinks, so security can sign off before models touch production data.

  • AI workloads sharing accounts with production data and no isolation boundary
  • GPU quotas and node pools sized by hand, with no guardrails or budgets
  • Model endpoints reachable without private networking or workload identity
  • No inference logging sinks or cost tags required before a model can deploy

Data your models and reviewers can trust

Data engineering & DataOps

Ingestion, dbt transformations, orchestration, and streaming, with data contracts, quality gates in CI, lineage, and freshness alerts. Training inputs, RAG corpora, and features become versioned, tested, and reproducible on the stack you already run. A reviewer can trace which data fed an answer.

  • Training data and RAG corpora with no versioning or lineage
  • Schema changes break features and models silently; no data contracts
  • No freshness or volume alerts, so customers find bad data first
  • Hand-run notebooks and cron jobs instead of orchestrated, CI-tested pipelines

One record, source to model

  1. Sourcesapps, events, files
  2. Ingestionbatch and streaming
  3. Transformdbt, orchestrated
  4. Quality gatecontracts, tests, freshness
  5. Features and RAG corpusversioned, with lineage
  6. Modeltraining and retrieval
  • Record meets its contract and flows on
  • Schema drift held at the gate, nothing downstream breaks

Platform baseline you can cost and defend

FinOps for AI

GPU and inference cost attribution per product and tenant, model routing, budgets and alerts. Cost-control levers designed into the architecture instead of chased in month-end reports. We codify environments in Terraform/Pulumi and enforce tagging.

  • GPU spend unattributed to product lines or customers
  • No budgets or alerts on inference routes, so overruns surface at month end
  • Idle GPU capacity with no scheduling or scale-to-zero policy
DevEx is scored in every assessment
Check your own stack in two minutes

The evidence pipeline

Evidence is a by-product of the pipeline, not a document sprint

Every commit runs through tests, scans and signing, a policy-checked Terraform apply, and a runtime that logs, meters and tags cost. What falls out is what security, procurement, NIS2 (Network and Information Security Directive 2), DORA (Digital Operational Resilience Act) and AI Act reviewers ask for.

Commit, then CI with tests, scan, SBOM (software bill of materials) and signing, then Terraform apply with policy check and drift detection, then runtime logs, metrics and cost tags. Each step fills the evidence packs for security review, procurement, NIS2 (Network and Information Security Directive 2), DORA (Digital Operational Resilience Act) and the EU AI Act. Technical evidence, not legal advice.

Infrastructure you control

EU-controlled by default

Your workloads run inside an EU boundary, with keys you hold, open components you can replace, and an exit you can execute. Sovereign clouds are options we build on, not partnerships we sell.

See the control boundary

Assess. Build. Operate.

Start with an assessment that gives you the straight answer and the evidence inventory. Build with a Production Implementation Sprint (one workload through review in 4–6 weeks) or the Private LLM Platform Package (an EU-controlled LLM platform). Add the Platform Operations Retainer only if usage and regulation demand it. Scope and price are agreed before kickoff.

Assess

Production AI Stack AssessmentFrom €4,50010 business days

Build, pick one

Production Implementation SprintFrom €18,0004–6 weeks

or

Private LLM Platform PackageFrom €24,0006–8 weeks
Full assessment fee credited toward your build

Operate

Platform Operations RetainerFrom €6,000/monthMonthlyoptional
Start here10 business days

Production AI Stack Assessment: Standard

From €4,500

One product team, one primary workload (AI or not), and a security, procurement or compliance review coming up next quarter

  • 6-dimension maturity heatmap, scored 1–5 with the evidence behind each score
  • EU AI Act system inventory + evidence inventory: what reviewers will ask for, and what is missing
  • FinOps opportunity register: where GPU and LLM spend goes, and what to cut first
  • Data & delivery review: pipelines, contracts, continuous integration and delivery (CI/CD), rollback
  • DevEx scorecard and a 90-day ranked backlog, infrastructure as code (IaC) and automation first
  • The straight answer: go, fix-first, or pause
12–15 business days

Production AI Stack Assessment: Regulated

From €8,500

Multiple teams, high-risk use cases (EU AI Act Annex III), or an active procurement, NIS2, DORA or regulator review

  • Everything in Standard
  • Multi-team interviews
  • Provider vs deployer role mapping for every system
  • AI agent and Model Context Protocol (MCP) readiness, with a tool-auth review
  • Deep automation & IaC drift assessment across environments
  • Board-ready executive readout for procurement and regulator conversations

Which assessment fits your scope?

Four answers, matched to the published tier criteria.

Teams involved
AI workloads in scope
High-risk use case (EU AI Act Annex III)
Procurement or regulator review

Recommended

Standard

From €4,500 · 10 business days

Fits one product team, one primary workload (AI or not), and a security, procurement or compliance review coming up next quarter.

Book stack assessment

Full assessment fee credited toward your build

Book a Production Implementation Sprint or the Private LLM Platform Package within 90 days and we credit 100% of the assessment fee.

Book stack assessmentSee the 10-day plan

After the assessment

  • Production Implementation Sprint

    From €18,000 · 4–6 weeks

    One workload through review and into production in 4–6 weeks, not two quarters, in code your team owns. Landing-zone modules, CI/CD with rollback and signed builds, data quality checks, observability and cost tags. An LLM gateway with eval gates where a model is involved. A focused build, not an open-ended platform programme.

    Request sprint scope

  • Private LLM Platform Package

    From €24,000 · 6–8 weeks

    An EU-controlled LLM platform: gateway, guardrails, audit logging and cost control. Routed to EU-hosted APIs, or to self-hosted open-weight models when residency or model control demands it. Retrieval-augmented generation (RAG) with evals in CI, and a full LLMOps lifecycle on infrastructure you control. Not a SaaS API wrapper project, and not a vague LLMOps retainer.

    Request private LLM scope

  • Platform Operations Retainer

    From €6,000/month · Monthly

    Keep the evidence current as usage and regulation grow. Monthly platform and architecture reviews, FinOps governance, reliability work, and upkeep of the inventory, logs and documentation your reviewers ask for. Optional, never required to start.

    Discuss retainer

What lands on your desk in 10 business days

Fixed scope, board-readable outputs, and an engineering backlog your team can execute in Terraform, CI/CD, GitOps and your data pipelines. Or we run the build for you. Every artefact is written to be forwarded to a reviewer.

Assessment clock: Standard

  1. Days 1–2Discovery & system inventory

  2. Days 3–5Production & automation assessment

  3. Days 6–8Risk & evidence mapping

  4. Days 9–10Delivery & readout

Lands on day 10

  • Executive readout (10–15 slides)
  • 6-dimension maturity heatmap
  • EU AI Act system inventory (filled)
  • Evidence inventory for security, procurement, NIS2, DORA, and EU AI Act reviews
  • DevOps, IaC & DevSecOps gap analysis
  • Data pipeline findings (contracts, lineage, freshness)
  • FinOps opportunity register
  • DevEx scorecard
  • 90-day ranked backlog with effort bands

The system inventory and evidence gap analysis are engineering artefacts. Legal classification, NIS2 and DORA scoping, and regulatory submissions stay with your counsel.

Regulated adds (from €8,500, 12–15 business days):

  • Multi-team interviews
  • Provider vs deployer role mapping for every system
  • AI agent and Model Context Protocol (MCP) readiness, with a tool-auth review
  • Deep automation & IaC drift assessment across environments
  • Board-ready executive readout for procurement and regulator conversations

Illustrative sample, not client data · 1-5 maturity scale

Every assessment includes a scored heatmap across six production dimensions.

  • Data & pipelines3/5

    Quality, lineage, data contracts, feature stores, and reproducible training inputs.

  • Deploy & automation2/5

    CI/CD, IaC, GitOps, rollback, environment parity, and delivery metrics.

  • LLM & agents2/5

    Gateways, eval gates, prompt versioning, tool auth, RAG quality.

  • FinOps2/5

    GPU and inference attribution, routing, and budget guardrails.

  • DevEx3/5

    Golden paths, inner-loop speed, observability, on-call readiness.

  • Governance2/5

    Logging, access, documentation, NIS2, DORA and EU AI Act technical evidence.

Start before you book

EU AI Act system inventory template

Pre-assessment worksheet: register systems, draft risk tiers, and map technical evidence gaps.

Download the EU AI Act inventory template
Book stack assessment

From first call to a system your team can run

Fixed-scope entry, engineering delivery, optional operate. No open-ended statement of work (SOW) trap.

  1. 01

    30-minute fit call

    Free, with an agenda. We confirm your stack, the review that is blocking you, and whether a fixed offer is the right next step. If it is not, we say so.

  2. 02

    Production AI Stack Assessment

    10 business days (12–15 for Regulated). Interviews and a system walkthrough, then the straight answer: go, fix-first or pause. With it: your 6-dimension maturity heatmap, FinOps opportunity register, evidence inventory, and 90-day ranked backlog.

  3. 03

    Production build

    A Production Implementation Sprint (one workload through review in 4–6 weeks) or the Private LLM Platform Package (an EU-controlled LLM platform in 6–8 weeks). Fixed price per agreed scope. Code lives in your repositories from day one.

  4. 04

    Operate & improve

    Optional Platform Operations Retainer: keep the evidence current as usage and regulation grow. Platform reviews, FinOps governance, and reliability hardening.

For your reviewers

What each reviewer receives from an engagement

Four cards, one per reviewer, listing the artefacts they get and what each one answers. Forward the card to the person who has to sign off.

Can this be rolled back, monitored, and traced?

Receives

  • DevOps, IaC & DevSecOps gap analysis: CI/CD, GitOps, drift, secrets, rollback, pipeline security
  • 6-dimension maturity heatmap with the evidence behind each score
  • Reliability risks: service-level objectives (SLOs), incident response, on-call readiness
  • 90-day ranked backlog with IaC, CI/CD and GitOps items first, with effort bands

Forward this to your CISO or information security officer.

Where is data processed, who holds the keys, and which data fed an answer?

Receives

  • EU AI Act system inventory: models, agents, pipelines, vendors, and where each runs
  • Environment map (dev, staging, production) with IaC coverage noted
  • Data pipeline findings: contracts, lineage, freshness
  • A data processing agreement (DPA) where we process personal data on your behalf

Forward this to your DPO.

What does this cost per product and tenant, and what is the engagement price?

Receives

  • FinOps opportunity register with estimated impact
  • FinOps scan: GPU, inference, and cloud attribution gaps by product and tenant
  • Fixed price per agreed scope, known before kickoff; assessment fee credited toward a build within 90 days

Forward this to your CFO or controller.

Who is the supplier, and what fills the questionnaire?

Receives

  • Supplier facts: legal entity, register court and number, address, from our Impressum
  • Evidence inventory: what security and procurement will ask, and what is missing
  • Executive readout (10–15 slides), board-readable
  • Code and documentation in your repositories from day one

Forward this to procurement.

Next step

Get the straight answer and your evidence inventory in 10 business days

Fixed price per agreed scope, from €4,500, ending in an honest go / fix-first / pause verdict and a 90-day ranked backlog. Book a Production Implementation Sprint or the Private LLM Platform Package within 90 days and we credit 100% of the assessment fee.

Assessment deliverables

  • 6-dimension maturity heatmap
  • FinOps opportunity register
  • EU AI Act system inventory
  • Evidence inventory for security, procurement and compliance reviews
  • DevEx scorecard
  • Data & delivery findings
  • 90-day ranked backlog (IaC & automation first)
  • Executive readout
Book stack assessmentBook a 30-minute fit call

You talk to the engineer who does the work.