AI Systems Design Architect - Assistant Vice President

What matters most

• Own the platform architecture

  The role owns end-to-end design for a major GenAI platform domain and shared AI services, including model integration, orchestration, APIs, data flows, and downstream interactions.

• Set reusable design standards

  Define reference architectures, integration contracts, and non-functional requirements; establish reusable patterns for RAG, model serving, prompt workflows, agentic workflows, and AI application integration.

• Build for production operation

  Design cloud-based solutions for interoperability, scalability, observability, security, and resilience; work with engineering and DevSecOps on releases, monitoring, and production readiness.

• Embed governance and controls

  Ensure designs address Responsible AI, model governance, data handling, compliance, testing, risk, and production controls.

• Guide decisions across teams

  Conduct architecture assessments, design reviews, and proof-of-concept evaluations; advise teams on trade-offs and standards while communicating with business, product, engineering, data science, platform, and security stakeholders.

Likely challenges to think through

• Standardization versus use-case flexibility

  Potential challenge: Shared RAG and agentic patterns must be reusable without forcing every use case into the same design. Prepare to explain when to use a reference pattern, when to allow a documented exception, and how to assess the trade-off.

• AI capability versus control

  Potential challenge: Model integration and data flows may create tension between useful AI behavior and governance, security, and compliance requirements. Prepare a design walkthrough showing where data handling, testing, risk controls, and Responsible AI considerations enter the architecture.

• Proof of concept versus production readiness

  Potential challenge: A promising AI proof of concept may not yet meet resilience, observability, or scalability requirements. Prepare to describe measurable non-functional requirements and the evidence needed before recommending production use.

Topics to prepare

• Enterprise AI architecture and non-functional requirements

  The posting calls for end-to-end design, reference architectures, integration contracts, and NFRs. Suggested revision exercise: take a hypothetical shared AI service and explain how security, resilience, scalability, performance, and maintainability would shape its architecture.

• RAG, LLM integration, and reusable AI patterns

  The role explicitly covers LLMs, ML systems, vector databases, RAG, model serving, prompt workflows, and agentic AI workflows. Suggested revision exercise: compare when each pattern is appropriate and identify its data-flow, testing, and operational trade-offs.

• Distributed integration architecture

  API architecture, microservices, event-driven architecture, orchestration, and downstream integrations are stated requirements. Suggested revision exercise: reason through synchronous versus event-driven integration for a hypothetical AI application, including failure handling and contract boundaries.

• Cloud-native design and operational readiness

  The posting requires solution design on at least one of AWS, Azure, or GCP, with understanding of compute, storage, networking, identity and access management, security, and AI/ML services. Suggested revision exercise: explain how these capabilities support an observable, scalable AI service without assuming a particular cloud is mandated.

• Responsible AI and production controls

  The architect must address model governance, data handling, compliance, testing, model risk, and operational controls. Suggested revision exercise: identify the design decisions and review questions these concerns raise for a hypothetical RAG workflow.

Skills to revise

• Draft an end-to-end architecture

  Suggested hands-on exercise: draw a hypothetical AI platform flow from API through orchestration and model integration to downstream systems; annotate trust boundaries and NFRs. This rehearses the end-to-end system design the role owns.

• Specify an integration contract

  Suggested hands-on exercise: write a sample API contract for an AI service, covering inputs, outputs, authentication expectations, errors, and versioning choices. The posting explicitly calls for APIs and integration contracts.

• Prototype and evaluate a RAG pattern

  Suggested hands-on exercise: build a small, hypothetical document-retrieval flow using a vector database and an LLM, then record retrieval-quality and data-handling limitations. RAG, vector databases, hands-on GenAI experience, and proof-of-concept evaluations are explicit in the posting.

• Run a technical design review

  Suggested hands-on exercise: critique a sample AI solution design against security, resiliency, observability, scalability, governance, and standards adherence; document decisions and trade-offs. The role reviews designs across teams and advises on architecture decisions.

• Present architecture to mixed stakeholders

  Suggested hands-on exercise: deliver a short technical explanation and a business-facing explanation of the same proposed design, then prepare answers to risk and delivery questions. The posting requires leading architecture discussions with technical and business stakeholders.

• Map the requirements to evidence

  The posting requires 8–14 years in relevant architecture roles and hands-on GenAI, LLM, RAG, vector database, cloud, and integration experience. Prepare a one-page evidence map with specific examples you can substantiate; mark any gaps rather than inventing examples.

• Build an end-to-end architecture walkthrough

  Choose a relevant project you can discuss. Prepare a diagram showing model integration, orchestration, APIs, data flows, downstream systems, integration contracts, and your design decisions—the areas this role would own.

• Practice a shared AI-service design exercise

  As interview preparation, sketch a reusable RAG or model-serving service with a prompt or agentic workflow. State assumptions, component boundaries, non-functional requirements, and trade-offs; the posting does not specify a particular design exercise.

• Explain cloud and integration choices

  Prepare a short decision record for one cloud platform you know: cover compute, storage, networking, identity, security, and AI/ML services, plus API, microservice, or event-driven integration choices. Relate those choices to scalability, interoperability, and observability.

• Prepare a production-controls review

  Create a checklist for data handling, Responsible AI, model governance, testing, risk, compliance, monitoring, resilience, and release readiness. These are stated responsibilities; use the checklist to explain how you would identify open decisions with security and DevSecOps partners.

• Rehearse an architecture review

  Select a design or proof of concept you can discuss and prepare a brief assessment: alternatives considered, evidence used, standards or NFR issues found, and how you communicated trade-offs to technical and business stakeholders.

• Draft questions about the platform domain

  Bring questions on the scope of shared AI services, existing reference architectures and integration contracts, how reviews work across pods, and which production controls need the most attention. The posting names these responsibilities but does not describe State Street’s current implementation.