Technical practice
How would you design a LangGraph workflow to troubleshoot a network problem using ReACT, dynamic tool selection, and self-correction?
What to coverThe posting calls for these agentic patterns in Marvis workflows. Suggested preparation: explain how you would represent state, select and validate tool calls, limit loops, and decide when the agent has enough evidence to respond.
How would you design and operate an MCP server that exposes domain-specific troubleshooting tools?
What to coverThe role includes tool schemas, skill playbooks, registry management, and server versioning, deployment, and monitoring. Suggested preparation: discuss schema clarity, error contracts, compatibility across versions, and how a playbook would guide tool use.
How would you make LLM API calls reliable in a production agent?
What to coverThe posting specifies structured outputs, streaming, function calling, prompt engineering, error handling, and cost optimization. Suggested preparation: compare validation and retry approaches for malformed outputs or failed calls, including their effects on latency and cost.
How would you tune retrieval for a troubleshooting RAG pipeline?
What to coverThe role covers chunking, vector-store search, hybrid retrieval, re-ranking, and relevance optimization. Suggested preparation: explain how you would compare retrieval configurations using representative troubleshooting queries and inspect failures where relevant evidence is missed.
When should an agent use long-term memory, and how would you evaluate whether it helps?
What to coverThe posting names semantic, episodic, and procedural memory as areas of work. Suggested preparation: distinguish what each could retain, then discuss relevance, stale information, and a comparison against an agent without memory.
How would you evaluate changes to a non-deterministic agentic workflow?
What to coverThe role calls for metrics, test harnesses, offline evaluations, A/B tests, and evaluation of skill and tool configurations. Suggested preparation: describe how you would measure troubleshooting quality alongside tool errors, latency, and cost, while accounting for variation between runs and limitations of LLM-as-judge.
How would you choose a statistical or ML method to investigate an apparent network anomaly?
What to coverThe posting requires foundations in anomaly detection, time-series analysis, clustering, causal inference, or related methods; networking expertise is a plus, not a requirement. Suggested preparation: explain what data and baseline you would need, how you would check false positives, and why an observed association would not by itself establish a cause.
How would you run agent-generated analysis and transformation logic in a sandbox?
What to coverThe role includes sandboxed execution for multi-tool scripts, aggregation, and visualization. Suggested preparation: discuss limits on execution and data access, validation of inputs and outputs, and how failures would be reported back to the workflow.
How would you ship a Python-backed agent service to Kubernetes?
What to coverThe posting specifies production Python, backend APIs, container images, CI/CD, Kubernetes workloads, health probes, and resource optimization. Suggested preparation: walk through an API boundary, automated tests and image versioning, then explain how probes and resource settings support a safe deployment.
How would you diagnose a production agent that has become slower, costlier, and less accurate?
What to coverThe role assigns ownership of tracing, logging, cost tracking, and performance monitoring for agentic workflows. Suggested preparation: show how you would inspect traces across reasoning steps, retrieval, and tool calls, isolate a regression, and verify a fix against quality as well as operational metrics.
Behavioural practice
Tell me about a time you translated a domain expert’s procedure into a technical workflow.
What to coverThe posting calls for collaboration with network engineers and product managers to turn troubleshooting playbooks into skills, tools, and data pipelines. Suggested preparation: use a real situation, your actions to resolve ambiguities with experts, and the result, without assuming networking experience is required.
Describe a time you chose an approach independently under technical uncertainty.
What to coverThe role calls for significant independent judgment on complex technical problems. Suggested preparation: give the situation, the alternatives and evidence behind your action, and the result or what you learned.
Tell me about a time you investigated a production AI or software reliability problem.
What to coverThe posting emphasizes reliable, observable production systems and troubleshooting failures. Suggested preparation: describe the situation, your diagnostic and communication actions, and the result, including how you checked that the issue was resolved.
Describe a time you helped another engineer improve their work.
What to coverThe role may include team leadership and mentoring. Suggested preparation: use a real situation, explain the specific guidance or feedback you gave, and state the resulting change without claiming formal management responsibility.
Tell me about a time you explained a complex technical trade-off to decision-makers.
What to coverThe posting prefers the ability to articulate technical designs to senior leadership. Suggested preparation: identify the audience and decision, describe how you presented alternatives and uncertainty, and explain the resulting decision or feedback.