Agentic Portfolio Workflow Deep Dive

The first half of an Applied AI Engineer screen is a project deep dive into an agentic portfolio workflow: model and provider choices, workflow behavior, ownership of endpoints, data access and entitlements, agent-orchestrator composition, and firsthand MCP server work. The interviewer then checks motivation and role alignment before moving to Python coding. Requirements Prepare a concrete walkthrough of an agentic portfolio workflow you personally built. Be ready to address: Which models the workflow uses and why. Whether it uses Anthropic or other model providers. What the AI workflow does end to end. Which parts you personally owned across endpoints, data access, and entitlements. Whether the multiple agents and orchestrator are independent services, tool-based components, or prompt-based components. Whether you have personally built an MCP server. Notes Ownership checks The questions separate hands-on implementation from high-level familiarity. Keep the boundary between your own work and the broader team's work explicit. Role alignment Expect direct questions about why you are changing jobs and whether you are comfortable with a role that leans toward back- or middle-office systems. Preparation Rehearse a five-minute architecture walkthrough that traces one request across the model provider, orchestrator, agents or tools, endpoints, and data or entitlement checks; label each component you personally implemented. Write and defend two decision records: one for the model or provider choice and one for the agent-orchestrator boundary. For each, state the constraint, chosen design, rejected alternative, trade-off, and failure mode.