How we work

Start small. Prove value. Scale.

Three ways to engage, one measure-first approach, and no long lock-ins.

01 · Engagement models

Three ways to engage.

2–3 weeks

Architecture Sprint

A fixed-scope assessment of your idea or existing system.

  • Stakeholder & data workshops
  • Target architecture & risks
  • Prioritised roadmap with estimates

Fixed price

Ongoing

Embedded Experts

AI engineers or a fractional architect inside your team.

  • Flexible capacity
  • Works in your tools & rituals
  • Scale up or down monthly

Time & materials

Need something in between, or a different setup? Engagements are tailored to you. Tell us what you need →

02 · Starting point

Find your starting point in 3 clicks.

Not sure which engagement fits? Answer three short questions and we'll suggest where to begin.

03 · Engagement process

From first call to handover.

Every step, how long it usually takes and which engagement it belongs to. Pick an engagement model to see only its steps.

  1. Free · 30 min

    Intro call

    We listen first: the problem, the goals, the constraints and whether we are the right fit. You leave with an honest first view of options, risks and cost.

    Architecture SprintBuild SquadEmbedded Experts

  2. 1–2 weeks

    Discovery deep dive

    Stakeholder interviews, a walk through current systems and data, security and compliance needs, and success metrics everyone agrees on.

    Architecture SprintBuild SquadNot in Embedded Experts

  3. Within a week

    Proposal & plan

    Scope, architecture outline, milestones, team and timeline in one short document, priced as a fixed price or time & materials.

    Architecture SprintBuild SquadEmbedded Experts

  4. Day 1

    Kick-off

    Access, environments and communication channels are set up, and we agree the demo rhythm and who signs off each milestone.

    Architecture SprintBuild SquadEmbedded Experts

  5. Agreed up front

    Milestones

    Each milestone has a clear deliverable and acceptance criteria, for example:

    • M1 · Data & architecture: data access confirmed, target architecture signed off
    • M2 · Working prototype: a thin end-to-end slice, measured against the eval set
    • M3 · Production-ready: hardened, monitored, tested and documented

    Not in Architecture SprintBuild SquadNot in Embedded Experts

  6. Weekly or bi-weekly demos

    Build in iterations

    A shared backlog, visible progress and a working demo every one or two weeks. Priorities can change between iterations; you always see what you are paying for.

    Not in Architecture SprintBuild SquadEmbedded Experts

  7. Continuous

    Internal testing

    Automated tests and evaluations for quality, safety, cost and latency run on every change, plus a security review before anything reaches your users.

    Not in Architecture SprintBuild SquadEmbedded Experts

  8. 1–2 weeks

    Stakeholder testing (UAT)

    You test with real users and real data. We collect the feedback in one place, so nothing is lost in email threads.

    Not in Architecture SprintBuild SquadNot in Embedded Experts

  9. Days, not months

    Adjust

    We prioritise the feedback with you, fix, and re-test against the same evals, so improvements are proven, not assumed.

    Not in Architecture SprintBuild SquadEmbedded Experts

  10. Per engagement

    Deliver

    What “done” means depends on the engagement:

    • Architecture Sprint: target architecture, risks and a prioritised, estimated roadmap
    • Build Squad: a PoC, pilot or production release
    • Embedded Experts: continuous releases inside your own delivery process

    Architecture SprintBuild SquadEmbedded Experts

  11. Final week

    Handover

    Code, docs, runbooks and training. Everything lives in your repositories and accounts; you own all of it, with no lock-in.

    Architecture SprintBuild SquadEmbedded Experts

  12. Optional · ongoing

    Operate & improve

    Monitoring, a support SLA, and model and prompt updates as providers, prices and your data change.

    Not in Architecture SprintBuild SquadEmbedded Experts

Your timeline or process looks different? We adapt to your governance, procurement and release cycles. Tell us how you work →

04 · Approach

Measure first. Build second. Scale last.

Step 01 · Discover

  • Stakeholder interviews & process walk-through
  • Data and systems audit
  • Success metrics and first eval set
  • Risk and compliance screening

Vendor-neutral

No reseller deals. We recommend what fits: open models, hosted APIs, or no AI at all.

Senior by default

The people you meet in the first call are the people doing the work.

You own everything

Code, prompts, models, evals and docs live in your repositories from day one.

05 · FAQ

Common questions.

We don't have an AI strategy yet. Is it too early?

No, that's the best time. An Architecture Sprint helps you avoid expensive dead ends and pick the use cases worth building first.

Can you work with our existing cloud and security policies?

Yes. We design within your constraints, AWS, Azure or GCP, private networking, data residency in the EU, and your identity and access model.

Will our data be used to train third-party models?

Not unless you decide so. We default to setups where your data stays in your tenant and provider data-retention is disabled.

Do you only work with large enterprises?

No. We work with startups, scale-ups and enterprises. Engagements are sized to the problem, not the logo.

Do you work with both hosted APIs and open-source models?

Yes. We are vendor-neutral: Anthropic, OpenAI, Google, Mistral and AWS/Azure/GCP model services, as well as open models (Llama, Qwen, Mistral, Gemma) self-hosted in your cloud. We choose based on your evals, data constraints and cost.

Can you build an MCP server for our product or internal systems?

Yes. We design agent-friendly tools, implement remote MCP servers with OAuth and scopes, and add a registry and gateway if you run many of them.

Which data platform or cloud should we choose?

It depends on your existing cloud, team skills and workloads, see the comparison on our Platforms & Cloud page. We are vendor-neutral and often run a short proof-of-concept on your real workloads before recommending one.

Can you migrate our AI workloads to another cloud provider?

Yes, for example Azure OpenAI to AWS Bedrock, Vertex AI to Azure, or APIs to self-hosted models. We add an abstraction layer, evaluate old and new side by side and cut over in phases with rollback.

How do contracts and billing work?

Fixed price for sprints, milestone-based for build projects, and monthly time & materials for embedded experts. B2B contracts across the EU and beyond.

Next step

Let's talk about your AI system.

A free 30-minute call with an Engagement Lead or AI Architect. You'll leave with a clearer view of options, risks and cost, whether or not we work together. Your case doesn't need to fit any box on this site; just tell us what you're facing.