AI Engineer – Agentic AI & Cloud Discovery

What matters most

• Deliver the assigned security workstream

  The role covers one of two workstreams: Agentic AI Forensics turns traces, prompts, tool calls and security events into explainable findings for analysts; Cloud Discovery classifies cloud resources and workloads to identify AI systems and infer purpose and risk at scale.

• Move from Python experiments to Go services

  The posting calls for prototyping prompting, RAG, structured extraction and classifiers in Python, then productionizing LLM-powered analysis and classification pipelines as Go services that meet defined accuracy targets.

• Measure and maintain model quality

  The engineer will define ground-truth datasets, evaluation metrics and regression suites, then monitor production quality and drift and address degradation.

• Make security results actionable

  The role involves working with security researchers to turn attack patterns and risk signals into detection and summarization logic, then integrating results with event, storage and UI layers for security teams.

• Use AI-assisted development with judgment

  AI coding assistants are a stated must-have. The posting says interviews will cover how candidates use these tools in development and validate their output; the role also calls for independent ownership in a remote-first team.

Likely challenges to think through

• Balancing explainability with coverage and scale

  Preparation hypothesis: forensics findings need to be understandable to analysts, while cloud discovery must classify varied resources at scale. Prepare to explain how you would choose evidence, structured outputs and review thresholds without treating a plausible LLM explanation as proof.

• Keeping evaluations representative

  Preparation hypothesis: ground truth and regression tests may be difficult to keep current as agent behavior, cloud inventories and risk signals change. Prepare a concrete approach to labeling, selecting error metrics, monitoring drift and deciding when degraded results need investigation.

• Hardening fast-moving AI prototypes

  Preparation hypothesis: moving Python experiments into Go services may create tension between iteration speed and reliable production behavior. Prepare to discuss how you would preserve measured quality while testing interfaces, failure handling and cloud-native deployment—and how you would validate AI-generated code rather than accept it unchecked.

Topics to prepare

• Agent architecture and failure modes

  The posting requires understanding of tool calling, MCP or similar protocols, multi-step planning, and common failure modes. Exercise: diagram a multi-step agent run, marking where a tool result or planning decision could produce an incorrect security finding.

• LLM analysis and classification design

  LLM APIs, prompting, structured outputs, and RAG are explicit requirements. Exercise: compare when retrieval, structured extraction, or an ML classifier would fit a security-event summary versus a cloud-resource classification, and explain how each output would be checked.

• Evaluation and production quality

  The role calls for ground truth, accuracy targets, regression suites, and monitoring for drift. Exercise: define a small labelled example set, suitable metrics, a human-review step, and a rule for investigating a drop in quality.

• Agentic AI Forensics workstream

  This workstream turns traces, prompts, tool calls, and security events into explainable findings for analysts. Exercise: trace how those inputs could support a finding, and identify which evidence an analyst would need to inspect. Treat the example as interview preparation, not a specified product design.

• Cloud Discovery workstream

  This workstream classifies resources and workloads to identify AI agents, model endpoints, and AI-enabled applications, then infers purpose and risk. Exercise: list the evidence needed to distinguish those categories and cases that should remain uncertain. Cloud APIs, IAM, and resource inventory are listed as preferred knowledge.

• Prototype-to-production architecture

  The posting pairs Python experimentation with Go services and cloud-native deployment. Exercise: sketch how a Python prototype becomes a Go-backed capability connected to event, storage, and UI layers; identify where evaluation and quality monitoring fit.

Skills to revise

• Prototype a Python pipeline

  Python prototyping of prompting, RAG, structured extraction, and ML classifiers is part of the job. Practice on a small illustrative set of agent events or cloud-resource descriptions; produce structured outputs and inspect ambiguous cases. This dataset is a practice suggestion, not an employer-provided dataset.

• Productionize an analysis path in Go

  Strong Go skills for production services are required. Practice implementing a small service that accepts an analysis request, validates a structured result, and handles an unsuccessful LLM response; explain what must change from the Python prototype.

• Build an evaluation regression suite

  The posting explicitly asks for ground-truth datasets, metrics, and regression suites. Practice labelling a few examples, recording expected outputs, running an offline comparison, and examining errors before accepting a change.

• Rehearse cloud-native deployment

  Experience with Docker, Kubernetes, and AWS, GCP, or Azure is required. Practice explaining or demonstrating how to package and deploy the Go service using Docker and Kubernetes on one of the listed cloud platforms, including how its quality would be monitored after release.

• Validate AI-assisted development

  Using an AI coding assistant is a must-have, and interviews will discuss how its output is validated. Practice using one listed assistant, such as Claude Code, Cursor, GitHub Copilot, or Codex, for a bounded code or test change; review the result, run tests, and explain what you corrected or rejected.

• Surface actionable security results

  The role integrates AI capabilities with event, storage, and UI layers and works with security researchers on detection and summarization logic. Practice turning a sample risk signal into a concise finding with supporting evidence and a structured result that those layers could consume. The sample format is a preparation suggestion, not a specified employer schema.

• Map evidence to the stated requirements

  Prepare a one-page map of relevant examples for Go, Python, production LLM/ML features, LLM APIs, RAG, structured outputs, evaluation, agent architectures and cloud deployment. Mark gaps honestly; the posting specifies 4+ years in software engineering and 2+ years shipping LLM/ML features.

• Rehearse a production project walkthrough

  If you have a relevant project, prepare a concise account of the Python experimentation, Go service design, your contribution, deployment, measured results and lessons learned. Explain how you worked independently and communicated decisions in a distributed team.

• Sketch a workstream-specific design

  As a practice exercise, choose agent-trace analysis that produces explainable analyst findings, or cloud-resource classification that identifies AI applications and security risk. Draw the inputs, LLM/ML pipeline, Go service and outputs; state how you would handle uncertain results.

• Build an evaluation plan

  For your design, outline ground-truth examples, accuracy metrics, human review, regression cases and production drift checks. Include likely failures, such as misleading tool-call traces or ambiguous cloud resources, and how you would detect degradation.

• Explain the production boundaries

  Prepare a diagram showing Docker and Kubernetes deployment on one cloud provider you know, plus connections to event, storage and UI layers. Be ready to discuss agent tool calling, MCP or similar protocols, and multi-step failure modes where relevant.

• Document AI-assisted development practice

  Prepare a specific example of using an AI coding assistant for implementation, testing or debugging, including what it produced, how you validated it and what you changed. The posting says interviewers will ask about this workflow.

• Write questions for the employer

  Ask which of the two workstreams the opening serves, how accuracy targets and ground truth are defined, how security researchers review findings, and how model drift is monitored. These are questions to clarify the role, not stated company processes.