ai-engineering-from-scratch/phases/17-infrastructure-and-production/01-managed-llm-platforms/quiz.json at main · rohitg00/ai-engineering-from-scratch · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
{
  "lesson": "01-managed-llm-platforms",
  "title": "Managed LLM Platforms — Bedrock, Vertex AI, Azure OpenAI",
  "questions": [
    {
      "stage": "pre",
      "question": "Before this lesson, which platform would you reach for first if a customer needed Claude, Llama, and Cohere behind one API?",
      "options": [
        "AWS Bedrock",
        "Vertex AI",
        "Direct Anthropic API",
        "Azure OpenAI Service"
      ],
      "correct": 0,
      "explanation": ""
    },
    {
      "stage": "pre",
      "question": "Which hyperscaler's bet is described as exclusive partnership rather than marketplace?",
      "options": [
        "Vertex AI",
        "AWS Bedrock",
        "OCI Generative AI",
        "Azure OpenAI"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "Roughly what is the measured median TTFT gap between Azure OpenAI (with PTUs) and Bedrock on-demand on Llama 3.1 405B equivalents?",
      "options": [
        "~250 ms",
        "~5 ms",
        "~100 ms",
        "~25 ms"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "Which Bedrock 2025 feature gives the cleanest per-product cost attribution natively in CloudWatch?",
      "options": [
        "Bedrock Agents",
        "Bedrock Knowledge Bases",
        "Application Inference Profiles",
        "Bedrock Guardrails"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "Azure PTUs typically break even versus on-demand at what sustained utilization band?",
      "options": [
        "40-60%",
        "80-95%",
        "5-10%",
        "20-30%"
      ],
      "correct": 0,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "Why is the lesson's recommended 2026 policy a two-provider minimum for product-critical LLM calls?",
      "options": [
        "Hyperscalers refuse to sign BAAs unless you also use a competitor",
        "Single-vendor pricing is always more expensive",
        "It is required by SOC 2 Type II",
        "Frontier model leadership rotates monthly, so single-vendor lock-in shuts you out of two-thirds of the frontier"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "Which statement best describes the FinOps surface across the three platforms?",
      "options": [
        "All three expose identical per-request attribution",
        "Bedrock is cleanest native, Vertex is most flexible via BigQuery, Azure is most opaque without instrumentation",
        "Vertex has no attribution surface at all",
        "Azure is cleanest native, Bedrock is opaque"
      ],
      "correct": 1,
      "explanation": ""
    }
  ]
}