Principal AI Architect

About us

Mainspring is a UK AI implementation firm helping engineering and manufacturing businesses put AI to work on their own terms. We build governed knowledge layers on each client's records, systems and approvals, and use them to bring scattered information together, automate repetitive technical work, and monitor machines and sites.

Our solutions are model-agnostic, cost-modelled before we build, and deployed where the client's data needs to sit. We are practical and evidence-led, and we are honest with clients when AI is not the answer.

Our team has founded and sold software companies, designed large-scale industrial machinery and managed manufacturing partners. Our projects span carbon removal, agriculture, field engineering and energy consultancy.

Joining Mainspring means working on real problems for real clients. You will work alongside engineers, consultants and compliance specialists, and your work will go live in the businesses we serve and make a measurable difference there.

The role

Before we quote a client, someone has to work out what we should actually build for them. How much foundation the problem needs. Whether their data, systems and sign-off processes will support it. What it will cost to run once it is live. And what evidence would prove it worked. That is this job.

Some clients need the lot: everything the organisation knows connected up, rules about who can see what and which sources can be trusted, enforcement every time AI is used, and a record you could put in front of an auditor. Others need a focused tool and barely any of that. Judging which is which is the hardest call we make. Build too much and we lose on price. Build too little and the client does not get what they came for.

Part of the job is telling people things they did not want to hear. That the honest answer is far smaller than they imagined. Occasionally that AI is not the right tool at all. We say that to clients, and we need someone who will say it internally too.

You also hold the few things that stay the same no matter how deep a build goes: how we prove a system works before it goes live, how we estimate what it will cost to run, and how we handle client data. And you look after the library of patterns our engineers reach for, the proven ways to do retrieval, permissions, audit trails and approvals. That library is there to save people re-solving solved problems, not to tell them how to work.

This is a hands-on job in a small team. You will build prototypes rather than write specifications from a distance, sit in client conversations where your credibility decides whether we win the work, and get your way through good arguments and working examples rather than through headcount.

What you'll be doing

Sit in opportunity reviews and discovery sessions, and work out what the client's systems, data and approvals will realistically support.
Decide how much foundation each engagement actually needs, and make the case when that is less than everyone expected.
Design the solution: where the information comes from, who can see what, how the workflow runs, what technology we use and where it gets deployed.
Estimate what it will cost to run before anyone puts a price on it, and improve the way we estimate.
Decide what counts as proof for each engagement, so the team knows what evidence the build has to produce.
Call out what cannot be done as imagined, early enough for the commercial team to do something about it.
Look after the pattern and component library, and decide what is good enough to go in it.
Hold the standards that apply to every build regardless of size: how we evaluate, how we estimate, how we handle data.
Set the model strategy so clients are not locked to one provider and can move if a better or cheaper option appears.
Design how our three deployment options work: public AI with proper controls, ring-fenced and private, or a mix of the two.
Be available to the engineers during the build, for the moments when reality does not match the design.
Work out the next stage when a client wants to extend, building on what is already there.
Raise the standard of engineering across the UK and Malaysia teams, through design reviews and working alongside people.

This role does not own any individual build, the budget, the client relationship, or the commercial negotiation.

Who we're looking for

Key competencies

You can judge how much to build, and you are willing to recommend less than the client was hoping for.
You get people to follow your lead through standards, prototypes and argument, not authority.
A client's IT or security lead will take you seriously in a room.
You still write and review production code. This is not a role that only produces diagrams.
You understand that a decision made before pricing is one we live with for the whole engagement.
You are realistic about cost and maintenance, and about what a small team can carry.
You want the engineers around you to get better, and you put time into that.

Essential

You have architected and shipped at least two production AI systems that real people relied on.
You have scoped work before it was priced, and then been held to your estimate.
You have built systems where permissions, provenance and auditability had to survive outside scrutiny.
You know how LLM systems fail, how to catch it, and how to prove reliability to someone who is sceptical.
Deep practical knowledge of retrieval, agent orchestration, tool and API integration, and managing context.
You can hold a technical conversation with a client's IT, security or engineering lead and keep their confidence.
Strong experience with Azure, AWS or Google Cloud, including containerisation and infrastructure-as-code.
Usually this comes with seven or more years in software engineering and at least two as an architect or technical lead. We care more about the evidence above than the number.

Desirable

You have worked in consultancy or client services, where scope comes before price and no two jobs are alike.
Experience in regulated or safety-critical work, and a working knowledge of UK data protection.
You have deployed AI into private, VPC or on-premises environments.
Experience running local or open-source models where volume makes cost the deciding factor.
Familiarity with industrial data: telemetry, historians, MES or engineering document sets.
You have defined responsible AI controls, or done red-teaming or threat modelling on AI systems.

What we offer

Meaningful work with real impact for established engineering businesses
A small, experienced team with a supportive and collaborative culture
Opportunities to learn as cutting-edge problems are solved
Flexible working that suits the role
Competitive salary and benefits

Our interview process

1Intro call30 minutes
2Technical interview60 minutes
3Team discussion45 minutes
4OfferWe aim to move quickly

How to apply

Send us your CV and a short note about a solution you scoped before it was priced. What you recommended, what you deliberately left out, and how it held up once it was built. Links to code, writing or talks are welcome.

If you do not meet every requirement but are excited by the role and bring key strengths, we encourage you to apply. We welcome applications from people of all backgrounds, and every application is reviewed by a member of our team.