Senior AI Ops Architect

What matters most

• Deliver measurable enterprise AIOps outcomes

  The posting calls for production-scale GenAI-powered AIOps that improves MTTR and supports automated incident resolution. Prepare to explain architectural decisions and how operational improvements were measured, without assuming a particular target.

• Architect GenAI retrieval and model workflows

  The role covers LLM solutions, RAG, vector search, and MLOps deployment workflows. Prepare to explain how retrieval, model deployment, and evaluation fit into an operational use case rather than presenting tools in isolation.

• Automate cloud-native infrastructure securely

  Responsibilities span AWS, Azure, GCP, infrastructure as code, Kubernetes, CI/CD, and DevSecOps. Prepare to discuss choices for scaling, deployment, and security automation across those layers.

• Connect observability to incident workflows

  The posting requires end-to-end observability and integration with ServiceNow and other ITSM tools. Prepare to trace how telemetry becomes a useful incident signal and then an appropriate workflow action.

• Provide architectural direction

  The architect is expected to guide leadership and engineering teams on AIOps and operational intelligence. Prepare to communicate design trade-offs and operational outcomes at both strategic and implementation levels.

Likely challenges to think through

• Speed versus control in incident automation

  A likely design challenge is reducing MTTR without allowing an incorrect AI-generated action to worsen an incident; this is a preparation hypothesis, not a reported EPAM problem. Prepare to discuss confidence thresholds, validation, and when to require human approval.

• Integration breadth versus architectural simplicity

  Supporting multiple cloud, observability, and ITSM technologies may complicate data flows and maintenance; this is an inference from the scope of the posting. Prepare to explain where common interfaces help and where platform-specific capabilities justify exceptions.

• Useful detection versus alert noise

  Combining anomaly detection, retrieval, and operational telemetry may make it hard to distinguish actionable incidents from noisy signals; this is a likely trade-off, not a confirmed issue. Prepare to discuss signal quality, evaluation against incident outcomes, and feedback into model or workflow improvements.

Topics to prepare

• Enterprise AIOps architecture and MTTR

  Posting: Design production-scale, Gen-AI-powered AIOps solutions that improve MTTR and automate incident resolution. Revision exercise: Sketch an incident-to-resolution architecture and explain where automation acts, where an operator intervenes, and how you would measure MTTR change.

• LLM and agentic-AI design choices

  Posting: Architect LLM solutions using Amazon Bedrock, Azure OpenAI, Vertex AI, Anthropic, LangChain, and Bedrock Agents; agentic AI is also named. Revision exercise: Compare two listed platform options for an incident-assistance use case, explaining orchestration and operational trade-offs.

• RAG and semantic retrieval

  Posting: Implement RAG, vector databases, and semantic search using PGVector, Elasticsearch, and Bedrock Knowledge Sources. Revision exercise: Diagram how an incident query retrieves relevant operational documents, then explain how you would check retrieval quality before using the answer.

• MLOps and operational ML

  Posting: Develop model deployment workflows with SageMaker or Azure ML and use clustering, topic modeling, and anomaly detection. Revision exercise: Outline a deployment and monitoring lifecycle for an anomaly detector, including how its alerts would reach incident operations.

• Cloud-native operations and security

  Posting: Design across AWS, Azure, and GCP; operate Kubernetes, CI/CD, security automation, Istio, KEDA, and listed serverless services. Revision exercise: Explain when you would use Kubernetes versus API Gateway, Lambda, and Step Functions in an AIOps design, and identify deployment and security controls.

• Observability and ITSM integration

  Posting: Implement end-to-end observability with tools including OpenTelemetry, DataDog, Dynatrace, and Splunk, and integrate incident workflows with ServiceNow, Jira Service Management, or Manage Engine. Revision exercise: Trace a signal from telemetry through incident creation, automated action, and resolution; identify what each system must exchange.

Skills to revise

• Architecture walkthrough

  Posting: The architect must deliver scalable enterprise AIOps solutions and guide leadership and engineering teams. Practice presenting a one-page incident-automation design, stating its components, failure points, and proposed MTTR measure. The presentation format is a preparation suggestion, not a posting requirement.

• RAG retrieval exercise

  Posting: Hands-on RAG and semantic search are required. Practice indexing a small set of sample runbooks with PGVector or Elasticsearch, querying for an incident symptom, and checking whether the retrieved passages support the response.

• Anomaly-detection workflow

  Posting: MLOps pipelines, deployment workflows, and anomaly detection are required. Practice specifying a small telemetry dataset, training an anomaly detector, recording its evaluation results, and outlining deployment through SageMaker or Azure ML.

• Infrastructure and deployment automation

  Posting: Terraform, CloudFormation, Helm, Python, shell scripting, Kubernetes, and CI/CD are named. Practice drafting infrastructure configuration and a Helm deployment for a small AIOps service, then walk through CI/CD checks, rollback, and security automation.

• Observability-to-incident integration

  Posting: End-to-end observability and ITSM integration are responsibilities. Practice mapping an OpenTelemetry signal to an alert and a ServiceNow incident, including the data needed to update and close the incident. Using a sample workflow is a preparation suggestion.

• Map experience to the stated requirements

  Prepare a one-page chronology of relevant architecture, cloud, SRE, and automation work. The posting specifies 19+ years in IT and 10+ years in those areas; distinguish documented experience from gaps.

• Build a production AIOps walkthrough

  Prepare a diagram and concise account of a GenAI-powered AIOps solution you delivered, if applicable: your decisions, production scope, incident workflow, and measured MTTR or automation outcomes. Use only results you can substantiate.

• Practice an incident-resolution design exercise

  Sketch a proposed enterprise flow from observability signals through anomaly detection and AI-assisted triage to an ITSM action. Be ready to explain scalability, security controls, human approval, and how you would measure outcomes; this is a preparation exercise, not a stated employer process.

• Explain your LLM and ML design choices

  Prepare a short comparison of relevant Bedrock, Azure OpenAI, or Vertex AI choices, plus a RAG design using semantic search and a vector store. Add how you would deploy and evaluate anomaly-detection models; focus on tools you can discuss credibly.

• Show cloud delivery trade-offs

  Create an architecture sketch covering infrastructure automation, Kubernetes and CI/CD, and a serverless component where appropriate. Explain choices among the posting’s cloud and tooling options rather than implying every tool belongs in one solution.

• Rehearse an architecture briefing

  Prepare a two-minute spoken summary and a one-page written brief for leadership and engineers. Explain operational value, delivery risks, and decisions clearly; the posting calls for thought leadership and written and spoken English at B2 level or higher.

• Prepare targeted employer questions

  Write questions about the current AIOps platform, available observability and ITSM integrations, production success measures, and the architect’s decision-making scope. These are questions to clarify, not assumptions about EPAM’s implementation.