ModulesSAI-220
SAI-220 table of contents
Model Serving, Routing, and Hardware
Select and operate model runtimes, routing patterns, hardware profiles, capacity controls, and reliability targets.
Learning outcomes
What you should be able to do
- Match runtimes and hardware to workloads
- Design governed model endpoints and routing
- Plan capacity, reliability, and failure behavior
Curriculum
Work through 4 sections in order.
The chapters are individually addressable documentation pages. You can link directly to a concept from another program, architecture decision, or implementation guide.
Serving foundations
Connect workload requirements to runtime and deployment boundaries.
Introduction to model serving
Translate workload behavior into governed runtime, endpoint, routing, hardware, and service requirements.
Deployment boundaries and patterns
Compare public cloud, private cloud, on-premises, hybrid, edge, restricted-network, and air-gapped patterns without treating one as automatically sovereign.
AI assets and attack surfaces
Inventory the data, models, prompts, artifacts, identities, tools, interfaces, infrastructure, and evidence that require protection.
Runtime, routing, and hardware
Design governed endpoints, routing policy, capacity envelopes, and hardware placement.
Runtime and endpoint architectures
Separate model runtime, API contract, policy enforcement, scheduling, and workload responsibilities.
Routing, capacity, and hardware
Design routing and hardware profiles from measured workload envelopes, isolation needs, and failure behavior.
Reliability and operations
Define release, degradation, recovery, observability, and controlled-change behavior.
Serving security, reliability, and operations
Operate model endpoints with explicit access, degradation, recovery, telemetry, and change controls.
Evaluation and release gates
Turn acceptance criteria into repeatable promotion decisions with recorded evidence and rollback conditions.
Capacity, reliability, and recovery
Define capacity envelopes, degradation behavior, service objectives, recovery targets, and tested restoration evidence.
Observability and operational evidence
Design signals that explain service health, AI behavior, policy outcomes, change, and incidents without leaking sensitive content.
Controlled change and versioning
Version complete AI behavior, assess change impact, approve promotion, support rollback, and retire superseded assets.
Decide, apply, and assess
Record the runtime decision and demonstrate it against measured requirements.
Architecture decision records
Record context, options, control consequences, evidence, dependencies, limitations, and reversal triggers for material decisions.
Model-serving design workshop
Create and test a runtime decision for a defined private AI workload.
SAI-220 knowledge check
Verify runtime, routing, hardware, reliability, evaluation, security, and operational decisions.
Practical completion package
- Serving workload profile
- Runtime and API contract decision
- Routing and hardware-capacity design
- Reliability and degradation plan
- Measured serving and release evidence
Current release boundary
This curriculum does not certify a runtime, accelerator, benchmark result, or production capacity. Fast-changing runtime commands and hardware compatibility belong in separately tested adapters with explicit versions and measurement conditions.