Generative AI Platform Manager, Vice President

What matters most

• Set Bedrock and Azure AI platform direction

  The role owns Bedrock strategy and Azure AI Foundry/OpenAI architecture, including model access, account or project design, capacity, and RAG and agent workloads. Prepare to explain architectural decisions across both clouds, not just individual model integrations.

• Lead the Databricks data and model platform

  The posting calls for Databricks strategy on AWS and Azure, Unity Catalog and Mosaic AI adoption, lakehouse RAG, and fine-tuning or pretraining pipelines. Focus preparation on how data governance, model serving, and workload design fit together.

• Create reusable, secure deployment paths

  The VP is expected to author CDK constructs, Bicep modules, and Terraform, and own Harness templates and pipelines. A key priority is making secure-by-default paths reusable across teams, with policy checks, scanning, model evaluation, and cost checks built into delivery.

• Own operational governance and economics

  Responsibilities include responsible AI controls, audit logging, guardrails, cross-platform observability, and token, capacity, and DBU cost controls. Prepare to discuss how these controls inform platform decisions and budget ownership.

• Scale adoption through technical leadership

  The role sets technical direction, mentors and recruits engineers, partners with Data, MLOps, Security, and Application leaders, and enables teams to ship at scale. It also calls for internal MCP servers and secure AI-assisted developer tooling, with productivity impact measured.

Likely challenges to think through

• Balance common patterns with cloud-specific controls

  Likely preparation hypothesis: reusable platform patterns may be difficult to reconcile with Bedrock multi-account and cross-region design, Azure quotas and data residency, and Databricks governance across both clouds. Prepare a decision framework for what to standardize and what must remain provider-specific.

• Make fast deployment compatible with safeguards

  Likely preparation hypothesis: golden paths intended to speed launches may face tension with approval gates, private access, security scanning, and model evaluation. Prepare to explain how automated defaults and clear exceptions could reduce friction without bypassing required controls.

• Manage quality, capacity, and cost together

  Likely preparation hypothesis: provisioned versus on-demand inference, Azure PTU versus pay-as-you-go capacity, and Databricks DBUs create competing cost and performance choices as usage grows. Prepare to describe workload-level measurements and thresholds that would guide routing, caching, capacity planning, and model selection.

Topics to prepare

• AWS Bedrock platform architecture

  The posting makes Bedrock a primary responsibility, including multi-account model governance, provisioned throughput, cross-region inference, and Knowledge Bases, Agents and Guardrails. For revision, sketch an architecture and explain when you would choose on-demand inference versus provisioned throughput.

• Azure AI Foundry and Azure OpenAI tradeoffs

  The role owns Foundry hubs and projects, model governance, quotas, and Azure OpenAI capacity planning. For revision, compare PTU and pay-as-you-go choices for a sample RAG workload, including content filters and data-residency constraints.

• Databricks data-centric AI architecture

  The posting names Unity Catalog, Mosaic AI, MLflow, Vector Search, Delta Lake, and fine-tuning and pretraining pipelines across AWS and Azure. For revision, diagram how governed data moves through a lakehouse RAG workflow to model serving, noting lineage and cluster-policy decisions.

• Security and responsible AI controls

  The role calls for private networking, Entra ID, customer-managed keys, audit logging, guardrails and red-teaming. For revision, walk through a proposed agent workload and identify where access controls, content safety, evaluation and audit records belong.

• FinOps and observability

  The posting explicitly calls for token, PTU and DBU governance, chargeback, capacity planning and monitoring across cloud and Databricks services. For revision, work through a hypothetical usage increase and explain which cost and operational signals would guide routing, caching or capacity changes.

• VP-level platform direction

  This is a hands-on leadership role with roadmap and budget ownership, executive communication, and mentoring senior engineers. For revision, outline a proposed sequence for delivering secure-by-default platform paths, then practice explaining its tradeoffs to engineering and executive audiences.

Skills to revise

• Prototype a Bedrock RAG path

  The posting requires Bedrock RAG and names OpenSearch Serverless, Aurora pgvector and Kendra. As a revision exercise, build or outline a small Python retrieval-and-generation flow using one named store, then show how you would evaluate answers, apply Guardrails and track token use.

• Configure an Azure RAG or agent workload

  Azure AI Search, prompt flow, Agent Service and Content Safety appear in the responsibilities. As a revision exercise, configure or draft a minimal workload using those components and document its Entra ID, private-networking and quota settings.

• Assemble a governed Databricks workflow

  The role covers Delta Lake, Unity Catalog, MLflow, Mosaic AI Model Serving and Vector Search. As a revision exercise, prepare a small lakehouse RAG workflow or implementation plan showing data permissions, index creation, model tracking and a DBU cost check.

• Author reusable multi-cloud infrastructure

  The posting asks for CDK L2/L3 constructs in TypeScript or Python, Bicep modules and Terraform, with secrets management and drift detection. As a revision exercise, draft a reusable AI-workload module interface and a secure deployment example for each named IaC approach.

• Build a Harness reference pipeline

  The role owns Harness YAML pipelines, templates and delegates integrated with IaC, GitOps and OPA policy checks. As a revision exercise, draft a pipeline that runs linting, security scanning, model evaluation and cost checks before an approval gate.

• Design a secure internal MCP server

  The posting calls for internal MCP servers and secure AI-assisted developer tooling in regulated environments. As a revision exercise, define a narrowly scoped MCP interface to a hypothetical enterprise system and specify its authentication, allowed actions and audit logging; treat the hypothetical system as practice, not a stated company project.

• Build a requirements evidence sheet

  Map the stated education, experience and technical requirements to specific, verifiable examples; distinguish required experience from preferred qualifications and mark anything not evidenced.

• Prepare a production platform walkthrough

  Choose one relevant project, if available, and explain its architecture, your decisions and hands-on contributions, deployment scale, incidents and measured outcomes. Be precise about which of Bedrock, Azure AI Foundry/OpenAI and Databricks it actually used.

• Sketch a multi-cloud design exercise

  Draft a proposed enterprise Gen AI architecture covering Bedrock account and model governance, Foundry hubs and quotas, and Databricks workspaces and Unity Catalog. Explain RAG or agent data flows and the trade-offs behind platform choices; present it as a preparation exercise, not State Street’s existing design.

• Show a secure deployment path

  Prepare a small CDK, Bicep or Terraform module outline and a Harness pipeline sketch with templates or delegates, secrets handling, OPA checks, scanning, model evaluation, approvals and drift detection. Identify what you have implemented versus what you would propose.

• Quantify governance and cost decisions

  Bring an example or worked scenario showing how you would test guardrails, audit logging and observability, then monitor token use, provisioned capacity, PTUs and DBUs. State assumptions and measurable cost or reliability outcomes only where evidence exists.

• Explain a developer-tooling approach

  Outline how an internal MCP server or AI coding assistant could be introduced with restricted access, secure usage guidance and a way to measure developer impact. Separate any prior deployment experience from this proposed approach.

• Rehearse the VP-level leadership narrative

  Prepare concise examples of technical direction, mentoring senior engineers, partnering with Data, MLOps, Security and application leaders, and owning a roadmap or budget. Clarify decisions made personally and how results were communicated to executives.

• Write targeted employer questions

  Ask how the role prioritizes Bedrock, Azure and Databricks; who owns platform standards and model governance; what adoption, safety and FinOps measures matter; and where teams most need reusable golden paths.