AI Architect

Role: AI Architect

Reports to: Chief AI Officer (line management)

Location in structure: AI/Data architecture area — embedded within the delivery area it serves, line-managing centrally to the Chief AI Officer

Experience: Senior — 6+ years in software, data, or platform engineering/architecture, with meaningful hands-on time on production AI systems, not just pilots

Direct reports: None. Influence comes from judgement and presence in the right conversations, not headcount.

AI decisions at Citation now cut across Product, Engineering, Security, Infrastructure, and the Business simultaneously — model selection, data architecture, cost, and risk are no longer separable concerns. Without a dedicated architectural owner, each initiative makes these calls independently: patterns diverge, risk goes unspotted until it's expensive, and nobody owns the AI-specific decisions that don't belong wholly to any one function.

This isn't a hypothetical gap. Much of this work is already happening informally inside Citation's AI delivery — reviewing partner Statements of Work, governing what goes through Code Factory, acting as the practical architectural voice on live builds. This role formalises that into a mandate with the standing and scope it needs.

The AI Architect is Citation's dedicated architect for the AI/Data area, sitting alongside the architects covering Human Resources, Business Systems, Health & Safety, eLearning, Verification, Certification, and Atlas Platform: embedded in the delivery area it serves day to day, but line-managing centrally to the Chief AI Officer so its calls hold across the business, not just the team it happens to sit nearest to.

The architecture hub owns target-state and standards across the whole architecture function; this role owns the AI-specific application of it, escalating decisions with consequences beyond AI/Data to the Architecture Review Board rather than deciding them alone.

The clearest sign this role is working: AI initiatives at Citation start well and stay on track architecturally. In practice that means:

• Established patterns are the default starting point for new builds, not something teams discover after the fact
• Design questions are resolved before Engineering starts building, not during or after
• Third-party Statements of Work are assessed architecturally before they're signed
• Model and hosting choices are made against a documented decision framework, not habit or vendor pressure
• Token spend and cost-per-outcome are tracked and explainable, not a surprise on the invoice
• Security is involved in every significant initiative from the start, not introduced at the end
• Leadership has a current, accurate view of AI architectural risk and direction, with no significant surprises

Architectural standards and patterns

Own the design patterns Citation builds its AI systems to: retrieval and grounding approaches for systems that need Citation's own knowledge rather than a model's general training, agent orchestration patterns and tool-calling conventions for multi-step and multi-agent work, prompt construction and guardrail design, and the routing logic that decides which model handles which step. Keep these current as the landscape moves, and write decisions down in a form Engineering and Product can actually build against — architecture decision records, not a slide deck.

Align standards to Citation's five-layer AI platform architecture.

Hold a documented, evidence-based framework for choosing between closed frontier models accessed through a provider's API (Anthropic, OpenAI, Google) and open-weight models run on Citation's…