Cloud AI services preview

Hyperscalers offer managed vision, speech, language, and custom training platforms. Cloud AI accelerates prototypes but requires governance: IAM, cost caps, data residency, and logging.

Service categories

• Pretrained APIs — OCR, translation, moderation
• Custom training — AutoML tabular, fine-tuning jobs
• Serving — managed endpoints with autoscaling
• Data labeling — workforce or assisted labeling tools

Integration pattern

# Pseudocode: call a hosted model API
import json

payload = {"inputs": "Summarize this ticket in one sentence."}
print(json.dumps(payload)[:50], "...")
# Real apps: auth headers, retries, timeouts, redact PII first

Practice: Optional Python pseudocode—pair with Python and data science tracks for hands-on depth.

Governance checklist

• Separate dev/staging/prod projects and keys
• Budget alerts on token and GPU spend
• Log prompts/responses with retention policy
• Region selection for data sovereignty

Important interview questions and answers

1. Q: Managed vs self-host?
   A: Managed reduces ops; self-host for air-gap or strict cost at scale.
2. Q: API key leak?
   A: Rotate keys, scope IAM, never commit secrets to git.

Self-check

1. List three cloud AI service categories.
2. Name two governance controls.

Pitfall: API keys in git—use secrets manager and separate dev/prod projects.