Choose the work before choosing the syllabus.
“Machine learning role” is not one job. Similar titles can test different skills because the day-to-day work, product stage, team and seniority are different.
Choose before revealing
Who is more likely to be tested deeply on deployment latency: a product Data Scientist or an ML Engineer owning online inference?
Five common role lenses
Data Scientist
Statistics, experimentation, SQL, modelling and business interpretation.
Machine Learning Engineer
Coding, ML pipelines, deployment, serving, monitoring and reliability.
Applied Scientist
Mathematical depth, modelling experiments, papers and novel approaches.
GenAI Engineer
LLMs, retrieval, evaluation, inference, guardrails and application engineering.
MLOps Engineer
Infrastructure, orchestration, CI/CD, observability and platform reliability.
Do not prepare equally for every role. Select one primary role and one adjacent role. The primary role controls the preparation budget.
A probable interview loop
Not every company uses every stage. The correct question is not “What do ML interviews ask?” It is “What evidence will this team need before trusting me with this work?”
Create a role hypothesis
Collect five representative jobs
Use roles you would actually accept, not arbitrary job descriptions.
Underline repeated responsibilities
Look for repeated verbs: build, analyse, deploy, experiment, optimize, partner, monitor.
Separate required evidence
Technical knowledge, ownership, scale, domain context and communication are different evidence categories.
Predict the interview
If the work says online recommendation, expect ranking metrics, retrieval, serving constraints and monitoring rather than only textbook classification.
Opening deliverable
Three target companies: __________, __________, __________
Most likely weak interview stage: __________