LIKELY QUESTIONS

- Walk me through a recent end-to-end analysis you owned on a large-scale system, from framing the question to influencing an engineering decision.
- How have you worked with telemetry or observability data from tools like Grafana, Elasticsearch, or Splunk to identify performance or reliability issues?
- This role involves high-dimensional workload data. How do you approach finding meaningful trends, separating signal from noise, and tying changes back to known events?
- Describe a time you uncovered a data quality or instrumentation problem during analysis. How did you debug it, validate the fix, and prevent recurrence?
- How would you design a lightweight classification or forecasting solution for datacenter workload behaviour that engineers would actually trust and use?
- Tell me about your experience using Python and JavaScript together to build analysis workflows, visualisations, or internal tools.
- How have you collaborated with OS, container, GPU, or infrastructure engineers when your analysis challenged assumptions or required changes from multiple teams?
- What experience do you have with Linux, HPC, or performance-sensitive environments, and how has that shaped the way you analyse systems data?

BEHAVIOURAL QUESTIONS

- Tell me about a time you had to make sense of ambiguous or incomplete telemetry data and still produce a useful recommendation.
Model approach:
- Situation: Analysis request had unclear scope and inconsistent instrumentation across services or clusters.
- Task: Produce a reliable view of workload behaviour and recommend next actions despite imperfect data.
- Action: Clarified the decision to support, mapped available signals, documented assumptions, triangulated across multiple sources, flagged confidence levels, and proposed instrumentation fixes.
- Result: Delivered actionable findings, avoided overclaiming, improved telemetry coverage, and enabled a follow-on performance or reliability improvement.

- Describe a time you influenced engineers or leadership using data visualisation and storytelling.
Model approach:
- Situation: Teams had competing theories about a performance regression or capacity issue.
- Task: Present complex, high-dimensional data in a way that drove alignment and action.
- Action: Reduced the problem to a few decision-relevant metrics, built clear visuals showing before/after and cohort differences, linked changes to events or releases, and tailored communication to technical and leadership audiences.
- Result: Stakeholders aligned on root cause or next experiment, work was prioritised, and measurable improvement followed.

- Tell me about a time you found that the original question being asked was not the right one.
Model approach:
- Situation: Team asked for a narrow analysis, but early exploration suggested a different underlying issue.
- Task: Reframe the problem without losing stakeholder trust or momentum.
- Action: Shared evidence early, proposed a sharper hypothesis, scoped a quick validation path, and documented why the revised question better matched the business or engineering decision.
- Result: Team avoided wasted effort, focused on the real bottleneck, and analysis led to a more valuable outcome.

- Give me an example of when you built something practical, not perfect, that became widely used by a team.
Model approach:
- Situation: Engineers needed faster access to workload insights, but existing analysis was manual or fragmented.
- Task: Create a lightweight tool, model, or dashboard embedded in the team workflow.
- Action: Prioritised the highest-value use case, kept the implementation simple, integrated with existing data sources and tools, added documentation, gathered user feedback, and iterated based on adoption.
- Result: Team saved time, analysis became repeatable, usage grew, and the tool informed regular debugging, planning, or forecasting decisions.

SMART QUESTIONS TO ASK

- What are the most important workload or infrastructure questions this team hopes this hire will answer in the first 6 to 12 months?
- How does this team currently partner with OS, container, GPU, and systems engineers, and where do analyses most often get stuck between teams?
- What telemetry stack and data sources are most central today, and where are the biggest gaps in instrumentation or data quality?
- When you say lightweight ML or DL inside existing workflows, what does successful production use look like here: offline decision support, near-real-time classification, forecasting for capacity, or something else?
- How do you measure success for this role: adoption of tools, quality of insights, performance improvements, forecasting accuracy, or influence on engineering roadmap decisions?

RED FLAGS TO WATCH FOR

- They cannot clearly explain what decisions this role supports, who the main stakeholders are, or what success looks like beyond "analyse data."
- They describe poor data quality, fragmented ownership, or missing telemetry but show no plan, investment, or appetite to improve the foundations.
- They emphasise advanced ML or flashy dashboards, but cannot point to real engineering adoption, operational impact, or cross-functional support for acting on findings.