AI Infrastructure Architect

What matters most

• Turn architecture into working AWS components

  The role owns hands-on engineering for production-quality LLM, GenAI and agentic AI components on AWS. It calls for code, design decisions and reusable patterns—not architecture documents alone.

• Build and evaluate enterprise RAG and agents

  The posting specifies Bedrock agent workflows, enterprise API integration, and retrieval over S3, OpenSearch and Knowledge Bases. It also calls for ingestion through retrieval, plus evaluation of accuracy, relevance, faithfulness and safety.

• Engineer security and operational controls

  Responsibilities include PII redaction, prompt-injection defenses, guardrails, authentication, authorization, lineage and audit logging. Components must be tested, observable and maintainable in collaboration with data, product and security stakeholders.

• Optimize the AI infrastructure stack

  The project-role description calls for custom AI infrastructure or hardware solutions, balancing performance, power consumption, cost and scalability, and advising on technology and vendor selection.

Likely challenges to think through

• Balancing infrastructure objectives with application delivery

  Preparation hypothesis: the broad infrastructure remit may need to be reconciled with the posting’s detailed AWS application-engineering duties. Prepare to explain how you would compare performance, power, cost and scale while still delivering a working component; the posting does not specify a particular hardware project.

• Improving retrieval quality without unbounded cost or latency

  Preparation hypothesis: parsing, chunking, embeddings and vector search create trade-offs between relevance, faithfulness, response time and operating cost. Prepare a concrete approach to measuring retrieval and answer quality, then tuning the pipeline against those measures.

• Giving agents useful access without weakening controls

  Preparation hypothesis: enterprise API integration and multi-agent workflows increase the difficulty of enforcing least-privilege access, PII protection and auditability. Prepare to describe control points, failure tests and observable evidence for a Bedrock-based workflow.

Topics to prepare

• AI infrastructure trade-offs

  The role explicitly covers infrastructure and hardware performance, power, cost, scalability and vendor selection. As revision practice, compare two hypothetical architectures against those criteria and explain what evidence would drive a selection; do not assume a particular hardware platform.

• LLM and agent architecture

  The posting requires prompts, tool calling, agents, orchestration, memory and context handling. Sketch a request flow from model invocation through a tool call and back, identifying state boundaries, failure paths and testing points.

• Retrieval architecture and evaluation

  The posting names ingestion, chunking, embeddings, vector search and evaluation for accuracy, relevance and faithfulness. Trace how a sample enterprise document reaches an answer, then identify where retrieval and answer quality can fail.

• Production AI security and governance

  PII redaction, prompt-injection defenses, guardrails, authentication, authorization, lineage and audit logging are explicit responsibilities. Threat-model a retrieval-and-tool workflow and map each risk to a control and a verification test.

• AWS service responsibilities

  Hands-on experience is requested across Bedrock, Agents/AgentCore, Knowledge Bases, Guardrails, Lambda, API Gateway, Step Functions, OpenSearch Serverless/Vector Engine, SageMaker and supporting security, monitoring and storage services. Draw a service-boundary diagram for an LLM workflow and justify what each selected service does, without assuming every service belongs in one design.

• Delivery and domain constraints

  The posting calls for client-facing, Git-based delivery and experience with AI/ML or data products in a listed industry. Practice explaining how domain data, process constraints and controls would affect a hypothetical design; treat the scenario as interview preparation, not a stated project.

Skills to revise

• Python LLM component implementation

  Hands-on Python coding and production-quality LLM components are required. Practice writing a small, testable Python component that prepares context, invokes a model interface and handles errors; explain its API contract and unit tests.

• AWS agent and API integration

  The role calls for Bedrock agent workflows and enterprise API integration through Lambda and API Gateway, with Step Functions among the required AWS experience. Practice outlining or prototyping a tool call with input validation, timeout handling and a tested failure response.

• Secure RAG pipeline construction

  The posting specifies secure RAG over S3, OpenSearch and Knowledge Bases. Practice defining an ingestion-to-retrieval workflow, including parsing, chunking, metadata and access checks, then test whether retrieved passages respect document permissions.

• Evaluation and safety testing

  Evaluation harnesses and tests for accuracy, relevance, faithfulness and safety are explicit. Build a small example test set with expected citations or grounded answers, add a prompt-injection case and record pass/fail criteria.

• Deployment and observability

  Automated testing, CI/CD, infrastructure-as-code, monitoring and runbooks are requested. Practice drafting a Git-based release checklist for one component with test gates, deployment configuration, CloudWatch signals, CloudTrail audit evidence and rollback steps.

• Map the stated requirements

  Prepare a one-page evidence map for Python, AWS, LLM engineering, industry-domain delivery, and technical leadership. Note that the posting specifies a minimum of 7.5 years of experience, a relevant bachelor’s degree, and 15 years of full-time education; leave any unsupported item unclaimed.

• Prepare a project walkthrough

  If you have a relevant AI/ML or data product example, outline the domain constraints, your own code and design decisions, stakeholder collaboration, testing, and operational results. Distinguish your contribution from the team’s.

• Practice an infrastructure trade-off exercise

  Sketch how you would evaluate an AI infrastructure stack for performance, power, cost, scalability, and vendor integration. This is preparation for the advertised architect remit, not an assumption about a specific client project.

• Design a secure AWS RAG pipeline

  Create a diagram and brief explanation covering ingestion, parsing, chunking, embeddings, retrieval, and an LLM response over enterprise content. Show where S3, OpenSearch, and Bedrock Knowledge Bases could fit, and explain your choices.

• Walk through an agent implementation

  Prepare a component-level example using Bedrock agents or AgentCore, tools, and orchestration, with an enterprise API connected through Lambda and API Gateway. Explain context handling, failure paths, and how you would test the workflow.

• Bring an evaluation and controls plan

  Draft sample tests for accuracy, relevance, faithfulness, and safety. Mark where PII redaction, prompt-injection defenses, guardrails, access controls, lineage, and audit logging would be applied.

• Prepare delivery evidence and employer questions

  Have concise examples of Python implementation, automated tests, CI/CD, observability, and a design or runbook artifact, if available. Ask how this role divides time between infrastructure and application engineering, and what production constraints shape client engagements.