Interviews
The people doing the work today, not the process as documented.
Home / AI Technology
AI Technology
Start with the problem. End with the right technology.
Three stages that turn an idea into working software, without committing to a stack too early.
- 2 wk
- Discovery to shortlist
- 1 doc
- Blueprint, signable
- 4 wk
- Blueprint to POC
- Yours
- Code and architecture
The failure mode
Most AI projects fail before the first line of code.
Not because the model was wrong. Because nobody agreed what problem it was solving, where the data lived, or who would run it afterwards.
-
Problem before stack.
The technology is the last decision, not the first.
-
Cost before build.
You see the year-two running figure before you commit.
-
Exit before entry.
If you leave, you keep the code and the architecture.
-
A real no.
If the answer is “not yet”, we say it and stop.
Where projects actually die: the wrong problem, data that is not ready, no owner after launch, a cost surprise. The model itself is rarely the problem.
Discovery
AI Discovery
Two weeks. We work out what is worth building, and what is not.
The data as it is
A look at residency, quality and gaps before anyone picks a model.
A baseline
Measurement now, so improvement can be proven later.
What you get
A ranked shortlist with effort, value and risk — plus an honest list of the ones to drop.
Blueprint
AI Blueprint
One document your CIO, CISO and finance lead can all sign.
What it contains
- Architecture and the data flow, drawn end to end
- Model choice, and why the alternatives were rejected
- Hosting, residency and the security review answers
- Build cost and the monthly running cost at year two
What you get
A costed, reviewable plan you own. Take it to us, to your own team, or to another vendor — it works either way.
Development
AI Development
POC, MVP, applications and integrations — inside your release process.
How we work
- In your repository, your branching model, your reviews
- An evaluation harness before the feature, not after
- Guardrails, logging and a documented failure mode
- Handover with runbooks, or we keep running it under SLA
What you get
Working software your team can maintain, with the tests that tell you when a model change breaks something.
Hosting
Where it runs afterwards.
Three options. We will tell you which fits your constraints.
Your account
Your cloud account
Your AWS, Azure or GCP tenancy. You hold the bill and the keys; we build and operate inside it.
When residency matters
Stellar Cascade
Our private cloud, single-tenant and in-country. Predictable monthly cost, fully managed.
See the platform →On-prem
Your own hardware
On-premise or colocated, including air-gapped. Slower to change, sometimes the only option that passes review.
Before you book
The questions that come up on every first call.
Do we have to buy all three stages?
No. Discovery and Blueprint are useful on their own, and plenty of clients take the Blueprint and build in-house. Each stage is priced and scoped separately, and you decide at the end of each one whether to continue.
What does a discovery cost?
A fixed fee agreed before it starts, based on the number of workflows and stakeholders involved. It is quoted in the proposal, not discovered halfway through. Development is either fixed-scope or a monthly team rate, depending on how well defined the work is.
Who owns what we build?
You do — code, architecture, models and documentation. There is no runtime licence to us and nothing that stops you taking it elsewhere. That is a deliberate constraint on how we design.
Will our data be used to train a model?
No. Your content is not used to train any shared or public model. Where a use case needs fine-tuning, the resulting weights belong to you and stay in your environment.
How fast can we start?
Discovery usually starts within two to three weeks of a signed proposal. If the timing is genuinely urgent we will say whether we can do it properly at that speed, rather than agreeing and under-delivering.
What if you decide we should not build it?
Then we say so, and the discovery output explains why. That has happened often enough that it is a stated outcome rather than an awkward surprise. You still have the assessment, which is usually worth having.
Book an AI Discovery.
Tell us the workflow you want to change. We will come back with scope, timing and a fixed fee.