Principal AI Architect

Title: Principal AI Architect

Location: United States (Santa Clara / New Jersey)

Work Model: Hybrid

Principal AI Architect, Financial Services (Americas)

Role Summary Financial services is one of the most demanding environments in which to deploy AI. The data is sensitive, the regulatory surface is wide, the legacy estate runs deep, and the consequences of a poorly governed AI system in production are real. What this environment asks for is genuine domain knowledge. Someone who knows how the business actually runs: the processes behind retail and commercial banking, how an account is opened and a customer onboarded, how a loan is originated and underwritten, how payments and transactions move, how a claim or a complaint is handled, and how risk, audit and compliance sit across all of it. That knowledge is what separates an AI design that survives contact with a bank from one that does not.

What You Will Do

Advisory Leadership and Client Engagement

• Our advisory portfolio is the job: The portfolio spans AI strategy and operating-model work, responsible AI and governance, production delivery and agentic engineering, and the assurance side covering adversarial testing and agent control. You are expected to know it cold and to carry it into the US financial services market as it stands, rather than reinventing an offer in front of a client.
• Defend the portfolio in the room: Financial services buyers push hard on scope, price and proof. You are the person who holds the line in front of a CIO, CTO, CISO

or CRO: why the engagement is shaped the way it is, what it costs, and what the client actually has at the end of it. Where a client challenges the approach, you answer with delivery evidence rather than by discounting the scope.

• Front the pre-sales conversation: Work alongside sellers, client partners and the wider business on US financial services pursuits. Qualify the opportunity, position the right entry point, shape the scope, and stand behind the proposal. Sellers do not run an AI pursuit alone, and on the most significant financial services pursuits you are the specialist who joins them.
• Lead the engagement into delivery: Advisory that stops at the recommendation is not what we sell. You lead the engagements you help win, from initial assessment through architecture and governance into build and then into production and run. Fixed scope, fixed price, a named deliverable at the end, built in the client environment on their cloud, with the reusable assets left behind for the client to keep.
• Grow the financial services position: Build the sector reference cases, work with the industry teams on the financial services domain content our engagements start from, and feed real delivery experience back into how our advisory work evolves.

Financial Services Architecture Leadership

• Domain-led solution design: Architect AI solutions that are grounded in US financial services reality. That means understanding the data models, operational constraints, regulatory requirements, and the organisational dynamics of US banks, insurers, broker-dealers, asset managers, and payments firms before walking into an architectural conversation.
• Regulatory and compliance architecture: Design AI systems that hold up under US federal and state supervision. Model risk, fair lending, financial crime, consumer protection, and data privacy obligations all place real constraints on how an AI system is built. You need to know which ones apply to a given use case and design for them from the start. Compliance is an architectural input, not an afterthought.
• Financial data standards: Work with the data standards that US financial institutions actually use: ISO 20022 for payments modernisation and SWIFT migration, FIX protocol for capital markets, Open Banking data sharing frameworks, and the reporting standards required by federal regulators. Design AI integration layers that operate against the core banking, insurance, and capital markets platforms that US institutions run on.
• Reference architecture ownership: Define and maintain the reference architectures behind our most common financial services AI use cases. These need to be opinionated, grounded in US regulatory requirements, and reusable across engagements rather than rebuilt each time.

Agentic AI in Financial Services Production

• Production-grade agentic systems: Design and govern multi-agent architectures that have been hardened for US financial services production environments. That means audit trails that stand up to internal audit and supervisory review, human-in-the-loop approval gates for high-risk decisions, real-time latency management for fraud and payments use cases, and operational monitoring that can detect model drift before it surfaces in an examination.
• AgentOps for regulated environments: Define AgentOps standards for AI systems operating under federal and state regulatory scrutiny. This covers agent decision trace logging sufficient for examiner review, hallucination monitoring and ground truth evaluation, model drift detection, cost tracking, and the change management and model validation processes that US regulated institutions are required to maintain.

• Core financial services process automation: Lead the architectural design of agentic AI across the workflows where it delivers most in the US market: onboarding and due diligence, financial crime investigation, lending and underwriting, trade and payment document processing, regulatory reporting, fraud case management, and complaints resolution. The focus throughout is on systems scaled and governed in production, not built in a sandbox.
• RAG architecture for regulated data: Define the standards for RAG pipelines that operate against US financial services data: transactional records, regulatory filings, customer data, and proprietary research. Designs must address data residency requirements, access controls aligned to need-to-know principles, audit logging for examiner access, and evaluation harnesses capable of detecting hallucination in contexts where the accuracy of the output has regulatory or legal consequences.

Multi-Agent Architecture Patterns

• Agent orchestration design: Define reference patterns for multi-agent financial services systems: hierarchical orchestration for complex compliance workflows such as financial crime investigation and case preparation, peer-to-peer agent collaboration for fraud and market surveillance, and sequential chains for document-heavy processes across lending, trade, and onboarding. Pattern selection should reflect the specific constraints of each client's technology environment and regulatory posture.
• Memory and state management: Architect memory systems for financial services agents: short-term session state for customer interactions, long-term knowledge via vector databases built on financial domain ontologies, and episodic audit memory for regulatory lookback and record-keeping requirements.
• Safety and guardrails: Design guardrail architectures for US financial services deployment covering prompt injection defence, sensitive financial data and PII masking, output validation against regulatory constraints, and escalation pathways for decisions that exceed defined risk thresholds. The failure modes of an agentic system in a US regulated institution need to be predictable, documented, and defensible under examination.

CXO Advisory

• Executive engagement: Lead strategic conversations with CIOs, CFOs, Chief Risk Officers, Heads of Compliance, and CDOs at major US financial institutions. These conversations need to operate simultaneously at two levels: the strategic discussion about AI adoption, regulatory positioning, and competitive differentiation, and the technical depth that earns credibility when the conversation turns to architecture, model governance, or implementation risk. Arriving with only the strategic view, or only the technical view, will not be enough.
• Partner and ecosystem influence: Engage at an architectural level with Anthropic and OpenAI, and with the hyperscalers that host them, Microsoft, Google and AWS. Know the strengths and the limits of each stack well enough to bring the right one into a client engagement, and to push back where the fit is not right.

Core Qualifications

• Financial services depth: 10 or more years working in or directly serving the US financial services sector in a technical or architecture role. Direct experience in at

least one of the following, with exposure to more than one an advantage: retail or commercial banking, capital markets, insurance, payments, or asset management. You understand how a supervised institution operates and what it takes to get an AI system through governance and into production there.

• Enterprise AI architecture: 8 to 10 or more years in technical leadership combining software engineering and enterprise-scale cloud architecture, with a proven track record designing and delivering multi-agent systems that have been scaled and operated in production under real governance pressure. Having done that inside a US financial institution is a strong advantage rather than a hard gate; other heavily regulated sectors count. Architecture documents and demo environments are not the same thing.
• Hands-on engineering: Still writing code and still building. You can pick up an agentic development SDK, stand up a working prototype against a real client problem in days rather than weeks, and put it in front of stakeholders to show what good looks like. Architects who have stopped building and now only produce slides and diagrams will not be effective in this role.
• Cloud expertise: Architectural expertise across at least one major cloud platform (Azure, GCP, or AWS) and hands-on familiarity with at least one other. Solid cloud-native engineering foundations: containers and Kubernetes, microservices and API design, event-driven patterns, CI/CD, and infrastructure as code. Awareness of US data sovereignty requirements and the GovCloud and financial services compliance configurations of each major provider.
• RAG and orchestration: Proven experience designing complex RAG pipelines. Hands-on development experience with at least one agentic development SDK or orchestration framework such as the OpenAI Agents SDK, OpenAI on Azure and Azure AI Agent Service, Google Agent Development Kit, LangGraph, or Semantic Kernel.
• Advisory and pre-sales: Track record selling and leading advisory-led engagements: qualifying opportunities, shaping fixed-scope offers, defending scope and price with a client, and then owning the delivery that follows. Comfortable working alongside sellers and account teams rather than waiting to be handed a signed statement of work.
• CXO communication: Sustained client-facing experience in a consulting, systems integrator, or vendor role, not solely internal or product work. The best candidates will have concrete examples of influencing an executive decision on AI strategy, model governance, or technology architecture on the basis of something they knew from actual delivery.

Preferred Qualifications

• AI governance frameworks: Experience applying recognised AI governance frameworks such as NIST AI RMF or ISO 42001 alongside a client's own model risk standards.
• Certifications: Professional-level cloud certifications such as Azure Solutions Architect Expert, AWS Solutions Architect Professional, or GCP Professional Cloud Architect, alongside AI credentials from Anthropic, OpenAI or NVIDIA.
• Thought leadership: Published research, conference speaking at industry events, or contributions to industry working groups relevant to AI in financial services.

What We Are Looking For

Someone who brings both halves: real financial services domain grounding and real AI architecture and engineering depth. Both are required. We are not expecting every item on the list above. A candidate who has taken agentic AI through model risk management inside a US regulated institution is the ideal case, but strong FS domain knowledge alongside production agentic AI delivered elsewhere in a regulated enterprise is equally credible.

Someone who builds. The fastest way to move a financial services client past skepticism is to show them something working against their own problem, and the people who can do that are the ones who never stopped being engineers. We want an architect who reaches for an agentic SDK and a laptop before reaching for a slide deck, and who can turn a whiteboard conversation into a running prototype in days.