UK-built · deployed worldwideOn-prem / air-gapped First build 4–12 weeksSource transferred to you Named human escalation

Expert AI engineering. AI systems designed, built and deployed in the UK.

We build production AI systems for enterprise organisations. From proprietary model training and bespoke AI SaaS to full AI architecture and engineering — built, evaluated and deployed into your infrastructure, with monitoring, governance and compliance from day one. We help organisations identify, optimise and deploy AI for innovation, efficiency, and growth.

01Capabilities

Three things we do properly.

Most agencies sell you the tool. We engineer the system around it — the retrieval, the evaluation, the escalation path, the audit trail, and the person who stays accountable when it gets something wrong.

01 / Engineering

Production systems, not prototypes

Agentic workflows, document pipelines and retrieval systems — architected, built, evaluated and deployed into your infrastructure with monitoring from the first day it runs.

  • Agentic orchestration
  • Document extraction at volume
  • Grounded retrieval
  • Evaluation harnesses
02 / Bespoke AI SaaS

Products we build, you own

Complete software products with AI at the core, deployed under your brand and owned outright. No per-seat meter, no platform lock-in, no renewal that doubles.

  • Multi-tenant architecture
  • Your brand and domain
  • Source transferred on completion
  • Built to scale past pilot
03 / Model Training

Custom-trained AI, inside your perimeter

A model trained and grounded on your organisation's own knowledge, running somewhere you control. It answers from what your business actually knows — and it structurally cannot leak.

  • Trained on your corpus
  • On-premise or air-gapped
  • Every answer cited to source
  • Nothing sent to third parties
02Model Training

Custom-trained AI, inside your perimeter.

Every general-purpose AI tool you subscribe to knows everything about the world and nothing about your business. Training on your own material inverts that — and never lets your knowledge leave the building.

We train and ground a model on your organisation's own material — your documents, your processes, your precedents, your tone of voice, your rules. Internally we just call it your business brain. It runs inside a boundary you control.

Ask it a question and it answers from what your business actually knows, then shows you the document it took the answer from. Because the model and the corpus both sit inside your estate, that knowledge is structurally incapable of reaching a third-party provider. It is not a policy promise that nobody sends data outside; it is an architecture in which there is nowhere outside to send it.

Grounded

Cited, not guessed

Answers retrieved from your corpus and traced to source. If it isn't in your material, it says so rather than inventing something plausible.

Contained

Deployed where you need it

On-premise, in your own private tenancy, or fully air-gapped with no outbound connection whatsoever.

Durable

Memory that survives people

When someone leaves, what they knew stays — searchable, attributable and current rather than lost with their inbox.

Reviewed

A person signs the ones that matter

Below a defined confidence floor the answer goes to a named human before it is used. Documented, logged, and reportable.

Your perimeter — nothing crosses this line

Layer 01 · CorpusYour organisation's knowledge

Contracts, procedures, precedents, correspondence, historic decisions — indexed, permissioned and version-controlled.

Layer 02 · RetrievalGrounded recall with citation

Every response traced to the source document and passage it came from.

Layer 03 · ModelTuned to your language

Trained on how your organisation actually writes, decides and operates — including how it says no.

Layer 04 · Human controlPermissions, logging, review

Role-based access, an immutable audit trail, and human sign-off wherever a decision carries consequence.

On-premise · private tenancy · fully air-gapped · How training works →

03Managed AI

Built is the easy half. Run is the hard half.

Most AI projects don't fail at launch. They fail four months later, quietly — when the model drifts, edge cases pile up in a queue nobody watches, and the cost per call triples without anyone noticing. Managed AI Services is the discipline that stops that.

Observability

Monitoring & evaluation

Continuous evaluation against a held-out set that reflects your real inputs, with alerts when quality moves rather than when it collapses. You hear it from us, not from a customer.

Model lifecycle

Drift detection & retraining

Input distributions shift as your business changes. We measure the shift, retrain against it, and version every change so you can roll back to a known-good state.

Economics

Cost & capacity control

Per-workflow cost tracking with hard ceilings, routing to smaller models wherever they perform identically, and a line-by-line monthly account.

Oversight

Human review queues

A defined confidence floor below which work goes to a person. We design the queue, define the escalation path, and report on what it caught — the catches are the evidence the control works.

Assurance

Incident response

A named engineer, a defined response window, a written post-incident record. AI systems fail silently and plausibly; the runbook accounts for that.

Accountability

Named accountable owner

One person owns each workflow, documented so it maps onto the governance you already have — an SM&CR responsibility map, a DPIA, or a board report.

04Technical depth

We build what we recommend.

A permanent core of UK-based LLM engineers, extended by a vetted network of specialists brought in by discipline — training infrastructure, retrieval, evaluation, security. You get the engineer who has done the specific thing before, not whoever was free that week.

Between them the team has shipped production AI systems across healthcare, retail, fintech and media. Everyone who works on your system writes code on it. We do not run an account layer between you and the people building the thing.

Selected production builds

Healthcare

HIPAA-compliant transcription

Regulated audio through speaker diarisation and multi-format export, behind scoped OAuth. Storage chosen for compliant handling rather than convenience.

Deepgram · Fly.io · Supabase · Next.js
Retail

In-store AI avatar

Camera detection driving a real-time greeting on a shop-floor screen over a WebSocket stream — containerised so the same build runs in every store.

HeyGen · Python · Docker · React · WebSocket
Fintech

Payment platform rebuild

Rebuilt with more than 500 tests, which surfaced around twenty billing defects in discount-stacking logic that had been live and unnoticed.

React · Supabase · Stripe Connect · Vitest
SaaS

Secure cloud storage MVP

Authentication, nested folders, in-browser file viewing and usage-based billing — specified, built and shipped in seven days.

Next.js · GCP · Stripe · Shadcn
Media

Three-stage image pipeline

A chained model pipeline — segment, inpaint, composite — rather than a single prompt and hope. Each stage independently testable.

Stable Diffusion · Next.js
Productivity

Presentation SaaS prototype

CSV-driven deck population with native .pptx export — the unglamorous format work that decides whether a tool gets used.

Next.js · React

What that engineering produced

20 hrs → 20 min

Weekly invoice processing, per person.

Global Logistics · 1200 staff · live in 6 weeks
3 days → 4 hrs

New client onboarding, automated end-to-end, including identity verification and source-of-funds evidence.

Accountancy practice · 1200 staff
60% faster

Client report production — the team now handles 40% more clients at the same headcount.

Marketing agency · 80 staff
SimpleAI have absolute competence in picking up your current tech stack with ease, or suggesting great alternatives. A fast and detailed AI development team with expert strategy, insight and delivery, and great communication.

The full technical stack, training and evaluation practice →

05Security & governance

AI you can put in front of your board, your auditor and your regulator.

Compliance documentation is produced as part of the build, not reconstructed afterwards when somebody asks. Every engagement ships with a data map, a DPIA and a named accountable owner — before the system processes anything real.

UK GDPR by design

Data mapping before build

We map personal data flows, processing locations, retention and access control before writing code. Minimisation is applied at the schema, not asserted in a policy.

Article 35

DPIA before deployment

Produced for any workflow touching personal data, and written to be read by your DPO and your auditor rather than filed. Residual risk stated plainly.

Article 22

The line we do not cross

No solely automated decision producing legal or similarly significant effects on a person. We build the evidence-gathering that gets a qualified human there faster.

Residency

UK by default

UK data centres, UK-incorporated processors, English law. Non-UK hosting is named explicitly and consented in writing, never assumed.

Secure development

NCSC and DSIT aligned

Model and dependency provenance, supply-chain controls, prompt-injection and data-poisoning threat modelling, and secrets handling that assumes the model is hostile.

Accountability

A name and a monthly report

Controls are designed against the SOC 2 Trust Services Criteria — we are not SOC 2 certified, and we will not imply that we are.

EU AI Act · Regulation (EU) 2024/1689 · status at September 2026 In force
2 Feb 2025 Prohibited practices enforceable — Article 5. Social scoring, untargeted facial scraping, emotion inference at work and in education. €35m / 7% turnover
2 Aug 2025 GPAI obligations, governance and the penalty regime in force. Chapters V, VII and XII. GPAI providers supervised by the European AI Office. €15m / 3% turnover
2 Aug 2026 Article 50 transparency obligations in force — unaffected by the Omnibus. Systems already on the market have until 2 December 2026 to meet Article 50(2). Live now
2 Dec 2027 Annex III high-risk obligations — deferred from 2 August 2026. By the Digital Omnibus on AI, in force 27 July 2026. Deferred
2 Aug 2028 Annex I high-risk — AI embedded in regulated products.Also deferred by the Omnibus. Deferred

Scope turns on role, not sector. Provider and deployer carry materially different obligations, and a UK organisation is in scope wherever a system is placed on the EU market or its output is used in the Union. We classify the role first, in writing, because every other obligation follows from it.

The deadline moved. The design requirement did not. A system built in 2026 without logging, traceability, human oversight and technical documentation will not spontaneously acquire them in December 2027. We build to the high-risk standard now, because retrofitting an audit trail into a system that was never designed to keep one is the single most expensive thing we get asked to do.

Trust, data residency, public sector procurement and what we won't build →

06Brand & growth

We have grown and monetised audiences from inside the platforms.

Senior content and audience leadership inside Google and Adobe. Delivery for Microsoft. Growth and monetisation across global publishers, B2B and B2C. We know how attention is built, measured and converted — because we did it at that scale before we built the AI systems that do it faster.

Audience

Growth systems, not campaigns

Editorial strategy, audience architecture and content engines designed to compound — the principles that scale a publisher, applied to a business that needs demand rather than traffic.

  • Audience & content architecture
  • Owned-channel growth
  • Measurement after attribution loss
Discovery

SEO, GEO and answer engines

Search is now one of several front doors. We build for the answer engines too — structured, citable content that language models retrieve and attribute.

  • Technical SEO & architecture
  • Generative engine optimisation
  • AI crawler access & policy
Monetisation

Revenue from attention

Turning an audience into a commercial asset — the discipline behind global publisher economics, applied to your funnel, pricing and lifetime value.

  • Funnel & conversion architecture
  • Brand identity & positioning
  • Lifecycle & retention
07Fractional

Senior capability, inside your team.

For organisations that need the expertise in the building rather than at the end of a statement of work. A fixed monthly commitment, working in your tools, in your stand-ups, against your roadmap.

Engineering

Fractional AI Engineer

An LLM engineer embedded in your team, building and maintaining your AI systems alongside your existing developers — and raising the standard of what your team can build without us.

  • Architecture & hands-on build
  • Code review & standards
  • Upskilling your engineers
Deployment

Forward Deployed LLM Engineer

Our most senior engagement. An engineer who sits with your operators, watches the actual work, and builds against what they observe rather than what a specification claims.

  • On-site discovery & build
  • Direct operator feedback loop
  • Documented handover
Growth

Fractional CMO

Marketing leadership for the AI era — owning brand, discovery and the content systems that generate demand, with the engineering capability to automate them rather than just recommend it.

  • AI brand & positioning
  • SEO and GEO strategy
  • Board-level reporting
08Command view

Our agents, working.

A live view of the fleet we operate — throughput, latency, extraction accuracy, and the escalations deliberately routed to a human. Amber is a person, every time it appears.

SimpleAI · Agent fleet · Command view
UK-South · UK-North · EU-West Streaming
Throughput / 24h
▲ 4% on trailing seven days
Median latency · ms
▼ 12ms week on week
Extraction accuracy · %
held-out evaluation set
Escalated to a human
by design, not failure
Throughput · trailing 24hdocs / 15 min
Fleet · 96 workerslive state
Idle Running With a person
Deployment regions · live loadrolling 60s
UK-South · on-prem0%
UK-North · tenancy0%
EU-West · tenancy0%
Review queue0%
Latency distributionp50 338 · p95 1,240
Escalation queuerouted to a named human
  • ◆Confidence below floor · contract clause extraction0.61 · 2m
  • ◆Ambiguous counterparty · onboarding evidence0.58 · 6m
  • ◆Conflicting source documents · policy answer0.54 · 11m
  • ◆Out-of-corpus question · no grounded answern/a · 14m

Every workflow has a confidence floor. Below it, work goes to a person — and we report what it caught. Illustrative telemetry — representative of deployed workloads. No client data, names or proprietary detail are shown or transmitted.

09Questions

Common questions.

The six we are asked on almost every first call.

What does SimpleAI do?

+

SimpleAI is a UK AI engineering agency. We design, build, evaluate and deploy production AI systems — agentic workflows, document pipelines, retrieval systems and complete AI products — into a client's own infrastructure. We work with enterprise organisations where the data is sensitive and the output has to withstand audit.

Where does our data live if you build us an AI system?

+

Wherever you need it to. We deploy in your own private tenancy, on-premise, or fully air-gapped with no outbound connection at all. UK data centres and UK-incorporated processors are the default; anything else is named explicitly and consented in writing.

Can an AI model be trained on our own documents?

+

Yes. We curate and de-duplicate your material, fine-tune on it, then align the model to how your organisation actually answers — inside a boundary you control. Every answer cites the source document it came from, and because the model and corpus both sit inside your estate, that knowledge cannot reach a third-party provider.

What will SimpleAI not automate?

+

Anything a regulator expects a named person to decide, citable output that reaches a client unverified, probabilistic checks that need to be exact such as conflict searches and sanctions screening, and anything whose reasoning cannot be explained after the fact. We build the work that leads up to a decision, not the decision itself.

Does the EU AI Act apply to us if we are a UK business?

+

Often, yes. The Act reaches any organisation placing an AI system on the EU market or whose system output is used in the Union, regardless of where the organisation is established. What matters most is your role — provider and deployer carry materially different obligations. Article 50 transparency duties have applied since 2 August 2026; the Annex III high-risk obligations were deferred to 2 December 2027 by the Digital Omnibus on AI, which came into force on 27 July 2026.

How do engagements usually start?

+

With a call, then a paid Discovery. Discovery produces a defined scope, wireframed flows, project timelines, budgets, a working interface preview and an implementation plan — artefacts you could hand to another builder if you chose to. A first build is typically four to twelve weeks from signed scope to live. How Discovery works →

Start here

We can help.

Tell us what's hard, expensive, or taking too long — and we will help you identify, optimise and deploy AI for innovation, efficiency and growth.