Tests your ability to reason about cloud compute selection, GPU vs CPU trade-offs, spot instance economics, and when to use managed vs custom training containers. Every question demands engineering judgment, not recall.

Deep-dives into AWS SageMaker features — Feature Store consistency, Pipelines caching, endpoint management — alongside Lambda and Serverless Inference cold starts, concurrency limits, and payload constraints. Tests production-grade AWS ML reasoning.

Tests your practical understanding of Vertex AI Pipelines, Hyperparameter Tuning, Feature Store, and BigQuery ML alongside Azure ML compute targets, pipelines data modes, and online endpoint auto-scaling. Cross-cloud reasoning for senior ML engineers.

Covers S3/GCS/Blob storage patterns for ML — Parquet vs CSV, small-file problems, versioning costs, and data lake design — alongside managed vector databases: Pinecone, pgvector, Vertex AI Vector Search, Weaviate hybrid search. Tests data-layer engineering judgment.

Covers LLM API usage patterns — token pricing, rate limits, vendor lock-in, prompt caching — cloud security for ML workloads (IAM, KMS, VPC, secrets), and cost optimization patterns including spot instance economics, reserved instances, and inference caching. The most production-critical cluster.

Simulates a real ML engineering screening interview. 10 questions covering cloud fundamentals, managed services, storage basics, and LLM API usage. Tests whether you can reason about cloud ML trade-offs — not just recall definitions. Every question has at least one believable trap.

Second easy-level mock interview. Focuses on managed service trade-offs, billing surprises, and IAM fundamentals. Every question is grounded in real production mistakes — cost overruns, misconfigurations, and common AWS/GCP/Azure gotchas that trip up juniors in live interviews.

Simulates a mid-level ML engineer interview. 12 questions spanning distributed training bottlenecks, pipeline caching behavior, vector DB query failure modes, and LLM token cost calculations. Requires multi-step reasoning — each question tests whether you understand the mechanism, not just the surface behavior.

Second medium mock. Tests your ability to diagnose production incidents: Feature Store consistency delays, S3 glob listing bottlenecks, Azure ML auto-scale cold starts, serverless endpoint compute limits, and Pinecone metadata filter degradation. Fresh question set — zero overlap with Mock Medium #1.

Simulates a senior ML engineer technical interview at a FAANG-tier company. 15 questions covering LLM memory budgets, HNSW vs IVFFlat drift under data growth, CUDA compute capability mismatches, KMS dual-policy access control, SCP permission ceilings, and GCP preemptible 24-hour hard limits. Each question requires multi-system reasoning.

Second hard mock — designed for staff/principal ML engineer and ML platform architect interviews. 15 questions across training memory analysis, SageMaker endpoint weight shift mechanics, Vertex AI artifact type compatibility bugs, Azure PTU overflow billing, PowerSGD gradient compression accuracy regression, Lambda SnapStart alternatives, and Cluster Autoscaler GPU node eviction. Zero overlap with Hard Mock #1.