Architect

Technical Skills Required

Practical experience building retrieval-augmented generation (RAG) or document-QA systems over large, frequently updated document sets.

Experience with LLM prompt design and evaluation for extraction and summarization tasks, including techniques to reduce hallucination and enforce citation of source text.

Ability to build lightweight pipelines for document version-diffing and change detection.

Familiarity with structured output formats (JSON schemas, tagged requirement registers) so extracted requirements can feed downstream tools or dashboards.

Working knowledge of data handling and confidentiality practices appropriate to regulated financial services content.

Payment scheme certification spans several independent regulatory and scheme bodies, each with its own documentation, update cadence, and terminology: card networks (Visa, Mastercard), card data security (PCI DSS), UK Open Banking (AIS, PIS, CBPII functional certificates), EU/UK payment services regulation (PSD2, PSR, FCA), and messaging standards (ISO 20022, SEPA). Keeping up with these manually is slow and error-prone. The AI engineers in this role will build and operate AI-assisted workflows (document ingestion, retrieval-augmented interpretation, change detection) that convert raw regulatory and scheme documentation into structured, reviewable guidance that downstream users can act on without needing to parse the originals themselves.

Key Responsibilities

Ingest and index regulatory and scheme source material (Visa/Mastercard mandate updates, PCI DSS standards, Open Banking functional certificate specifications, FCA/PSR publications, ISO 20022 message guides) into a retrieval system that stays current as sources are revised.

Build and maintain AI/LLM-based pipelines that extract certification requirements, deadlines, and scope changes from these sources, and that flag material changes between versions.

Translate extracted requirements into structured, plain-language summaries (requirement, applicability, deadline, evidence needed, owner) for consumption by compliance, product, and engineering teams who do not read the source documents directly.

Validate AI-generated interpretations against the source text before publication; maintain a human-review checkpoint for anything that affects certification scope, deadlines, or risk posture.