Stage 10: Production engineering
Decision framework, MLOps, serving, cost and latency, GPUs, monitoring, safety, reliability and the data flywheel.
Summary, rules and context: AI Engineering Guide, Stage 10.
Learn
Curriculum lessons, in reading order.
- Stage 10 — Production AI Systems: Overview
- Lesson 10.1 — Production Mindset and Decision Framework
- Lesson 10.2 — Data Pipelines and MLOps
- Lesson 10.3 — Serving Architectures
- Lesson 10.4 — Monitoring, Drift, and Cost
- Lesson 10.5 — Safety, Privacy, and Human-in-the-Loop
- Lesson 10.6 — Reliability and Incident Response
- Stage 10 — Exercises and “Ready for Stage 11” Checklist
Apply
Field-manual scenarios: implementation steps, an example, trade-offs, production notes, a checklist and metrics.
1-serving-cost-latency/
- Production → Cost levers
- Production → Latency levers
- Production → Self-hosted serving (vLLM/TGI, quantisation, GPU sizing)
- Production → Capacity & limits
- Production → Versioning & rollout
- Production → Environments
2-gpu/
2-gpu/details/
- 01 — Mental model: what GPU resize actually is
- 02 — Decision tree: whether to resize, and which axis
- 03 — Four axes: S, H, V, W
- 04 — Three resources and the sizing math
- 05 — Workload pools: never resize one fleet
- 06 — Considerations behind every resize
- 07 — Cloud and Kubernetes mechanics
- 08 — Serving knobs that must change with the GPU
- 09 — Trade-off matrices
- 10 — Production cutover playbook
- 11 — Examples (sizing, axis choice, K8s)
- 12 — Worked patterns for a prod AI app
- 13 — Finer details that bite in production
- 14 — Prod checklist and metrics
3-deployment/
4-observability-reliability/
5-guardrails/
- Testing & reliability → Input guardrails
- Testing & reliability → Output guardrails
- Testing & reliability → Prompt-injection defence
- Testing & reliability → Excessive agency (least privilege)
6-security-privacy-governance/
- Security → Authentication & authorization (permission-filtered retrieval)
- Security → Data residency & retention
- Security → PII (detect, mask, reversible mapping)
- Security → Auditability
- Security → Model & supply-chain hygiene
- Security → Threat checklist (OWASP LLM Top 10)
7-metrics-failure-modes/
8-data-flywheel/
Previous: Stage 9: Evaluation · Next: Stage 11: Architecture patterns