AI for UK Banking · Compliance-Ready · No Vendor Lock-in

Deterministic AI for banks
that cannot afford to hallucinate.

Production-grade AI systems for UK banking and lending. Full audit trail for your regulators. Code your team owns outright.

Book a free 45-minute call View reference architectures
£4,620
analyst cost recovered per night by AML triage engine
90%
token cost reduction via prompt caching on 50-page documents
0
hallucinated financial figures across all production deployments

Anthropic Partner Network

Registered tier

Claude Certified Architect (CCA)

Exam scheduled

Registered in England and Wales, Co. No. 15487606

Founded Feb 2024

Open Source

github.com/TechIfat

Built with
Anthropic Claude API
Model Context Protocol (MCP)
Pydantic · FastAPI
UK FCA / PRA Compliance
DevSecOps · HITL
Ifat Noreen, Principal Agentic AI Architect
AMD Research · 9+ yrs Regulated Finance AI Enterprise AI DevSecOps

About the founder

Built by an engineer.
Designed for compliance.

I am Ifat Noreen, founder of ShiftAi Systems Ltd. Before building AI systems I spent 9 years designing high-performance computing infrastructure at AMD Research, where the standard was not good enough but provably correct. I apply that same standard to every AI system I build for banks and lenders.

We do not sell software licences or retainer contracts. We embed within your team, build the system alongside your engineers, and hand it over completely. Your compliance team gets a system they can explain to the FCA. Your engineering team gets code they understand and can maintain.

Connect on LinkedIn View GitHub repos

Architecture Teardown

From 4 hours per credit application
to 4 minutes.

Technical Insight: Eliminating Hallucination Risk

Large Language Models are excellent at semantic extraction, but they cannot do math reliably. In this build, we completely locked Claude out of arithmetic. Claude extracts the raw figures from scanned balance sheets into strict JSON, then passes those variables to our deterministic Python Model Context Protocol (MCP) server. The Python server calculates DSCR and LTV, returning mathematically perfect results. The FCA gets an auditable process; the bank gets zero hallucination risk.

The Pipeline Flow

1. Ingestion
Unstructured PDFs
→
2. Sanitisation
Local PII Masking
→
3. Extraction
Claude Native Tool Use
→
4. Calculation
Deterministic MCP Server
→
5. Sign-off
Human-in-the-Loop
Read the full technical teardown on GitHub

Measured outcomes

What our reference architectures
actually achieve.

These are measured results from ShiftAi's own open-source reference builds, the same architecture patterns we deploy for clients. Client-specific results will be added here as engagements complete.

£4,620per night recovered
Analyst cost recovered by AML batch triage engine, every single night
500+
Transactions evaluated per nightly batch run at 50% compute cost discount
12 wk
From blank repo to production-grade platform, fully documented
100%
Deterministic JSON output: zero hallucinated entities or financial figures
Every design decision is documented and reviewable. Your CTO, CISO, and compliance team receive a full record of what was built, why each architectural choice was made, and how risk was controlled at every layer.
Request Full Brief →

Calculate your manual processing cost.

Estimate how much your compliance or underwriting team spends on manual data extraction before any decisions are made.

Manual Document Checks per Month 1,000 checks
Current Estimated Annual Cost
£300,000
ShiftAi Estimated Annual Savings
£282,000
*Based on an industry average of 45 minutes (£25 internal cost) per manual check vs ShiftAi's API & compute cost.

How we work

Three tiers. No retainers.
You own everything.

Tier 0 is a free discovery call. Tier 1 unblocks your risk committee. Tier 2 builds the system. Tier 3 keeps it running. Start at Tier 0. Stop whenever the case is not proven.

TIER 0 · FREE

45-Minute Discovery Call

A free video call on Zoom or Teams with your risk and engineering leads. We map your current document workflows, identify where AI errors would cause regulatory damage, and tell you honestly whether an AI Readiness Assessment is the right next step.

  • Video call on Zoom or Teams, whichever you prefer
  • High-level document workflow mapped
  • Immediate FCA/PRA risk points identified
  • Clear, honest recommendation on next steps
Book your free call
TIER 1

AI Readiness and Governance Assessment

A 2-3 week paid engagement mapping your current AI posture against FCA/PRA requirements. Deliverables include:

  • FCA/PRA compliance alignment matrix
  • Data PII leakage vulnerability check
  • Model hallucination risk audit
  • Fully scoped architecture proposal and fixed-price roadmap

Your CTO, CRO, and CISO leave with a document they can act on.

From £5,000. Fully credited against your Tier 2 build if you proceed within 30 days.
TIER 2

Architecture Sprint or POC

A 4-week contracted proof of concept or an 8-12 week full architecture sprint, scoped at the assessment stage. Deliverables include:

  • Working prototype demonstrated by week 2
  • Production-ready codebase with full test suite
  • Architecture Decision Records for every design choice
  • Complete handover documentation and operating runbook

No black boxes. No vendor dependency.

Scoped and priced at assessment stage
TIER 3

Sustained Adoption Programme

Once Claude is in production, we stand up your AI Centre of Excellence. Deliverables include:

  • Monthly AI Centre of Excellence sessions
  • Prompt library and governance framework
  • Token usage and cost monitoring dashboard
  • Quarterly model risk review against FCA/PRA guidance

Full IP transfer at handover. You retain complete control.

Ongoing retainer. Priced per engagement.

What we build

Seven ways we help UK financial services
deploy Claude safely.

Five open-source reference architectures plus two new service offerings. Every build targets a specific operational cost or regulatory risk in UK banking and lending.

01 /

Commercial Credit Underwriting

Reduces document extraction from hours to minutes. Reads scanned PDFs and balance sheets directly, routes all financial ratio calculations to a verified Python server, and holds every decision for human review before it is logged.

View codebase & ADRs
02 /

AML False-Positive Triage Engine

Clears 90%+ of overnight AML alerts automatically, leaving your compliance analysts to focus on the cases that genuinely need them. Every cleared alert includes a documented reason your team can review and your regulator can inspect.

View codebase & ADRs
03 /

Regulatory Gap Analyzer

Cuts regulatory gap analysis from months to minutes. Reads new FCA or PRA policy updates alongside your internal policies and produces a structured report with exact citations from both documents. No paraphrasing. No invented references.

View codebase & ADRs
04 /

Legal Redliner

Reduces commercial loan covenant review cycles from days to minutes. Two AI agents review each clause against your lending policy, challenge each other's reasoning, and only output a final draft when every clause meets the standard. Your lawyers review the output, not the raw documents.

View codebase & ADRs
05 / Featured capstone

UBO Corporate Unraveler

Reduces KYC beneficial ownership investigations from hours to seconds. The system traces corporate ownership chains through Companies House automatically, layer by layer, until it finds a human. It never invents an entity it cannot verify. Every step is logged for your audit trail.

View codebase & ADRs
06 /

AI Readiness and Governance Assessment

A 2-3 week paid engagement delivering an FCA/PRA compliance matrix, PII vulnerability check, hallucination risk audit, and a fixed-price Tier 2 build proposal. The £5,000 assessment fee is fully credited against your build if you proceed within 30 days.

Book an assessment
07 /

Credit Risk Memo POC

A 4-week contracted proof of concept. Credit analysts upload financial statements, news reports, and company filings. Claude generates a draft Credit Approval Memo following your bank's exact template. Reduces memo preparation from 10 hours to under 30 minutes. Converts directly to a full Architecture Sprint once ROI is demonstrated.

Book a POC sprint

Start here

One free call to find out
if this is worth pursuing.

45 minutes on Zoom or Teams with your risk and engineering leads. We map your current document workflows, flag where AI errors would cause the most regulatory damage, and tell you honestly whether an AI Readiness Assessment is the right next step. If there is no fit, we will say so.

Book your free call Free. 45 minutes. No obligation. hello@shiftaiconsulting.co.uk

Frequently asked questions

Common questions,
straight answers.

Security and architecture

No. The system runs on your own Anthropic API account and your own infrastructure, not ours, so ShiftAi never holds your data at all. Under Anthropic's standard commercial terms, API inputs and outputs are not used to train its models. Where your organisation requires a Zero Data Retention agreement with Anthropic directly, we can build to that requirement and confirm it as part of the AI Readiness Assessment.
We use a strict DevSecAI architecture. The AI operates in a sandboxed environment where read and write permissions are physically separated. Any action that alters a database, sends an email to a client, or approves a credit line requires a deterministic Human-in-the-Loop (HITL) interrupt. The AI can draft the decision, but a human must click approve.
We do not allow the LLM to do mathematics. We use the LLM purely for semantic extraction and orchestration. It extracts the raw figures from the document and passes them to a deterministic, tested Python Model Context Protocol (MCP) server to calculate DSCR, LTV, or affordability. The arithmetic is deterministic and independently testable, and the LLM simply formats the result for the report.
Because we build using abstracted frameworks like the Model Context Protocol (MCP), your system architecture is completely decoupled from the underlying LLM. If a model is deprecated, we update the API pointer, run your automated evaluation suite (LLM-as-a-judge) to ensure output quality remains identical, and push the update seamlessly.

Process and pricing

45 minutes on Zoom or Teams with your risk and engineering leads. We map your current document workflow, flag the two or three points where AI errors would cause the most regulatory or financial damage, and tell you honestly whether a paid AI Readiness Assessment is the right next step. No slide deck, no sales pitch. If there is no fit, we will say so on the call and that is the end of it. No fee, no obligation.
You keep the deliverables regardless: the data lineage map, the risk register, and the gap analysis are yours to use however you like. If we conclude AI is not the right fit for the workflow assessed, the report says so directly rather than recommending a build to generate fees. If you are not satisfied with the quality of the report itself, we will revise it based on your feedback at no extra cost.
The AI Readiness Assessment is a written report, so it does not depend on ongoing availability once delivered. For an Architecture Sprint, all code, architecture decisions, and documentation live in a shared GitHub repository from week one, with your engineering team given read access immediately, not just at handover. Every Friday you receive a written progress update. The project is never dependent on information that exists only in one person's head.
Yes. Every build engagement ends with a complete handover: the GitHub repository transferred to your organisation, production-ready code, Architecture Decision Records for every design choice, an operating runbook, and a knowledge transfer session with your engineering team. You own the code, the architecture, and the IP outright. We do not operate a SaaS model and there is no ongoing licence fee for the system itself.
Regulators do not pre-approve specific AI tools, but they do expect firms to demonstrate governance: human sign-off on consequential decisions, a full audit trail explaining why a decision was made, and a named individual accountable under SMCR. Every system we build is designed around those three requirements from the outset. The AI Readiness Assessment includes a gap analysis against current FCA and PRA AI guidance, written so your compliance team can present it to your board or your regulator directly.
The AI Readiness Assessment is fixed at £5,000 with no further commitment, and fully credited against a build if you proceed within 30 days. If you proceed to an Architecture Sprint, cost and timeline are agreed and fixed before any build work starts, based on the scope set out in the assessment report. Any request outside that agreed scope goes through a written change proposal, which you approve before any additional work begins or any additional cost is incurred.
The free call is a single 45-minute session. The AI Readiness Assessment takes 2-3 weeks. A full Architecture Sprint is typically 8-12 weeks from kickoff to handover, depending on scope. We have delivered production-grade architectures in 12 weeks from a blank repository.
Yes. Our reference architectures are built on the Anthropic API and Model Context Protocol (MCP). The patterns are stack-agnostic. We have built MCP adapters for legacy core banking APIs. If your systems expose an API, we can build a governed, auditable agentic layer on top of them.