Quiz

2948 questions of the kind that actually get asked — in interviews, in architecture review boards, and by the person who has to run the thing at 3 AM. Every answer states the trade-off rather than the slogan, and says when the obvious choice is the wrong one.

All areas2948 Architecture Fundamentals91 Distributed Systems101 Data Architecture90 Cloud Architecture87 Networking96 API & Integration Architecture87 Reliability & Resilience99 Observability92 Performance & Capacity Engineering100 Security Architecture105 Cost Architecture & FinOps102 Business Architecture103 Architecture Communication102 Enterprise Architecture100 Legacy Modernization102 AI-Era Architecture106 Software Architecture & Engineering93 Architecture Patterns94 Architecture Decision-Making101 The Architect's Meta-Skills102 Delivery & Release Engineering103 Platform Engineering & Developer Experience102 Testing & Quality Architecture102 Data Platform Architecture98 Streaming & Real-Time Data103 Data Governance & Semantics91 Frontend & Experience Architecture101 Edge, Mobile & IoT98 Regulatory & Data Protection Architecture99 Assurance, Audit & Model Risk98

Difficulty Any beginner intermediate advanced

6 questions in ML Platform.

Quiz

A machine-learning platform must ingest telemetry from many concurrent training runs. What are the workload's distinguishing characteristics?

A model performs well in evaluation and poorly in production. What are the likely causes?

A recommendation platform's ML systems span data pipelines, training, evaluation and online serving. Where should the boundaries be, and what causes the most costly class of bug?

A training run on thousands of GPUs loses several nodes mid-run. How should checkpointing frequency, elastic training, straggler detection and scheduling minimise wasted compute, and what does each checkpoint cost?

An interviewer says - we have about 30 models in production at a short-video platform at ByteDance scale and every team trains and deploys its own way. What do you centralise first, and what do you deliberately leave alone?

NVIDIA's NCCL implements the same all-reduce with several algorithms - ring, tree, and switch-assisted variants over NVLink and NVSwitch - and chooses between them at runtime. Why does one collective operation need several algorithms, and what does that force a cluster scheduler to know?