AI & Machine Learning Company in Leeds

Whether the model is clinical or commercial changes everything about the estimate: a scoped AI proof of concept at Leeds rates, a custom production ML model (fraud scoring, churn prediction, demand forecasting, document extraction), a clinical AI build with MHRA SaMD classification, DCB0129 clinical safety case, and UKCA marking via an approved body, or LLM and RAG systems at First Direct, Yorkshire Bank, or Leeds Building Society scale. Scope drivers include data audit, a baseline model, a hosted demo, MLOps, monitoring, a model card, the SaMD class, the volume of clinical evidence required, and FCA Consumer Duty foreseeable harm analysis. Leeds rates sit roughly twenty-five percent below London, fixed fee.

Yes. We have delivered AI work into NHS trust supplier ecosystems, and our clinical AI engagements ship with the artefacts NHS Digital and MHRA expect. Deliverables include DCB0129 manufacturer clinical safety case (compulsory under the Health and Social Care Act 2012 for software placed on the market for health and care use), DCB0160 deployer clinical safety case support, MHRA Software as a Medical Device classification analysis (Class I, IIa, IIb, III) under the UK MDR 2002 as amended, NHS Digital Technology Assessment Criteria (DTAC) version 1.0 readiness across clinical safety, data protection, technical security, interoperability, and usability, and UKCA marking route planning via an approved body where SaMD class demands it. We have patterned releases after the imaging AI deployments Leeds Teaching Hospitals has run at scale.

DTAC v1.0 is NHS England's baseline assessment for digital health technologies being commissioned across NHS trusts. It covers five domains: clinical safety (DCB0129 and DCB0160), data protection (UK GDPR DPIA, ICO registration, Data Security and Protection Toolkit submission), technical security (Cyber Essentials Plus, ISO 27001 alignment, penetration testing), interoperability (HL7 FHIR UK Core, SNOMED CT, DM+D where prescribing), and usability and accessibility (WCAG 2.1 AA minimum, NHS Service Manual alignment). Our Leeds clinical AI builds produce a DTAC-aligned evidence pack from kickoff, not retrofitted at procurement gate. We have walked Leeds Teaching Hospitals and other NHS trust procurement teams through this exact pack.

FCA Consumer Duty (Policy Statement PS22/9) has been in force from 31 July 2023 for new and existing products open to sale, and from 31 July 2024 for closed products. It requires firms to deliver good outcomes for retail customers and proactively avoid foreseeable harm. For AI in retail banking this means three concrete things: foreseeable harm analysis on every model materially affecting customer outcomes (credit decisioning, pricing, fraud-driven account restriction, complaint triage), monitoring across protected characteristics with documented intervention thresholds, and board-level evidence the firm is acting on the four outcomes (products and services, price and value, consumer understanding, consumer support). Our Leeds retail banking AI engagements deliver these as standard, with the Consumer Duty Champion (typically the senior NED) getting an evidence pack at each major release.

Yes. RAG systems, internal copilots, document extraction pipelines, and agent workflows are standard for Leeds financial and health clients. For First Direct, Yorkshire Bank, Virgin Money, and Leeds Building Society we stay inside UK regions (Azure OpenAI UK South, Bedrock eu-west-2, or self-hosted Llama 3 on UK GPU instances), apply Bank of England SS1/23 model risk controls where applicable, and produce challenger model documentation internal audit needs. For NHS clinical RAG (clinician-facing summarisation, EHR query, clinical guideline retrieval) we keep PHI in-region, document every retrieval and generation event for audit, apply red-teaming for clinical safety failure modes, and ship with explainability dashboards clinicians actually use. Prompt injection defences, PII redaction, and human-in-the-loop gates are mandatory in regulated builds.

Bias and fairness audits cover the protected characteristics under the Equality Act 2010 (age, disability, gender reassignment, marriage and civil partnership, pregnancy and maternity, race, religion or belief, sex, sexual orientation) plus socioeconomic variables relevant to FCA Consumer Duty foreseeable harm analysis. Our methodology covers data distribution analysis at training time, group fairness metrics (demographic parity, equalised odds, predictive parity, calibration) computed per protected group, individual fairness checks via counterfactual analysis, model behaviour audits with synthetic edge cases, and a documented mitigation plan where harm is identified. Where the protected characteristic is not directly observed (race is rarely captured in UK financial datasets) we apply geographic and name-based inference under strict ICO guidance and document the limitations explicitly.

A clinical AI build varies sharply with MHRA SaMD class. Class I SaMD (low risk, self-certification under the UK MDR) typically runs sixteen to twenty-six weeks. Class IIa SaMD with UKCA marking via an approved body adds clinical evaluation, technical documentation under MDR Annex II equivalents, post-market surveillance, and approved body assessment, taking total project to nine to eighteen months. Class IIb and III SaMD push past two years and require clinical investigation. Non-SaMD clinical decision support (informing rather than driving clinical decisions, with the clinician unambiguously in the loop) can ship in fourteen to twenty-two weeks with a DCB0129 clinical safety case but no UKCA route. We will tell you up front which bucket your problem sits in, and we will pull in clinical safety officers and approved body liaison as required.

Three reasons. First, our delivery model puts senior ML engineers and clinical safety-literate consultants on every project from Edmonton and Chandigarh hubs, not graduate consultants billed at partner rates. Second, fixed-fee pricing protects you from the day-rate inflation London AI consultancies and Big Four advisory have publicly admitted across 2023 and 2024. Third, we ship runnable model code, MLOps pipelines, and the regulatory artefact pack (DCB0129, DTAC, MHRA classification, FCA Consumer Duty foreseeable harm analysis) that NHS Digital, MHRA, FCA, and ICO reviewers actually accept. Leeds CTOs at First Direct, NHS trusts, and Yorkshire Bank scale have moved budget from London consultancies to our model on this exact comparison.