Cloud Pro: AWS AI and Machine Learning: Choose, Compare and Launch

What's inside

1. Level 1: The AWS AI and ML Map on One PageFree
   Learners place every AWS AI and machine learning service into six groups: generative AI platforms and models, agents, search and retrieval, ready-made AI APIs, machine learning platforms, and accelerators. For each service they record its current name, former names (for example SageMaker becoming SageMaker AI) and its lifecycle status as of October 2026.
2. Level 2: The Standard AI Workload ProfileFree
   Learners build the reusable workload profile that every later service choice is scored against. It covers task type, buy/adapt/train, data sensitivity and residency, latency and throughput, volume and unit costs, model choice and lock-in, evaluation and safety, MLOps maturity, accelerators, team skills and cost model.
3. Level 3: Bedrock Models and the Runtime APIs
   Learners see what Amazon Bedrock is and how to reach its foundation models through the runtime. They compare the Converse and InvokeModel APIs, model access, cross-Region inference profiles and Regional availability, and make the smallest working call from the CLI.
4. Level 4: Bedrock Pricing: On-Demand, Batch, Provisioned
   Learners work out what a Bedrock workload costs under on-demand token pricing, batch inference and provisioned throughput, and which levers reduce the bill: prompt caching, model choice and right-sizing prompts.
5. Level 5: Bedrock Knowledge Bases
   Learners use the nine-part pattern on Bedrock Knowledge Bases: managed ingestion, chunking, embeddings and vector store options for retrieval-augmented generation. The lesson covers the limits that bite, what it costs, and a minimal Knowledge Base launched from S3 and cleaned up afterwards.
6. Level 6: Bedrock Agents
   Learners use the nine-part pattern on Amazon Bedrock Agents: action groups, Lambda-backed tools, Knowledge Base attachment and orchestration. They also see how the current Agents differ from Agents Classic, which has been closed to new customers since 30 July 2026.
7. Level 7: Bedrock Guardrails
   Learners configure Amazon Bedrock Guardrails: content filters, denied topics, word filters, PII handling and contextual grounding checks. They attach guardrails to model calls and check what each one adds to cost and latency.
8. Level 8: AgentCore Runtime, Gateway and Identity
   Learners deploy an agent on Amazon Bedrock AgentCore Runtime. They expose tools through AgentCore Gateway and secure inbound and outbound access with AgentCore Identity, including VPC egress, and compare the cost and operating model with Bedrock Agents.
9. Level 9: AgentCore Memory, Tools, Policy and Evaluations
   Learners add the remaining AgentCore building blocks to a running agent: Memory, the built-in Code Interpreter and Browser tools, Observability, Policy and Evaluations. They finish with a pros and cons view of choosing AgentCore.
10. Level 10: Amazon Q Business in Brief
    Learners review what Amazon Q Business offers as a managed enterprise assistant: connectors, permissions-aware answers and per-user pricing. They see what its closure to new customers on 30 July 2026 means for new and existing deployments, and which alternatives AWS points to.
11. Level 11: SageMaker AI Training Jobs
    Learners run managed training on Amazon SageMaker AI: built-in algorithms and bring-your-own containers, instance choice, Managed Spot Training, checkpointing and distributed training. They learn the quota limits that bite, and that features such as Debugger have been closed to new customers since 30 July 2026.
12. Level 12: SageMaker AI Inference Options
    Learners compare SageMaker AI's inference options against latency, traffic shape and cost: real-time endpoints, serverless inference, asynchronous inference, batch transform and multi-model or inference-component endpoints. They also check what happens to monitoring now that Model Monitor is in maintenance.
13. Level 13: SageMaker HyperPod
    Learners decide when large-scale training or fine-tuning justifies SageMaker HyperPod. The lesson covers resilient clusters with Slurm or Amazon EKS orchestration, automatic node recovery, capacity planning and the costs of a cluster that is always on.
14. Level 14: SageMaker JumpStart
    Learners use SageMaker JumpStart to deploy and fine-tune open and proprietary foundation models on their own SageMaker endpoints. They compare this with calling the same kind of model through Bedrock on cost, control and operating effort.
15. Level 15: SageMaker Unified Studio in Brief
    Learners see where SageMaker Unified Studio fits: a single environment for data, analytics and AI work with domains, projects and governed data access. They separate it from the SageMaker AI service and from features now in maintenance, such as Studio Lab and Clarify.
16. Level 16: Trainium and Inferentia in Brief
    Learners look at AWS Trainium and Inferentia accelerators: the instance families, the Neuron SDK, what it takes to port a model to them, and when their price-performance beats GPU instances enough to justify the porting effort.
17. Level 17: Amazon Rekognition
    Learners use the nine-part pattern on Amazon Rekognition image and video analysis. It covers labels, faces, text and content moderation, per-image and per-minute pricing, and the April 2026 move of Streaming Events and Batch Image Content Moderation to maintenance.
18. Level 18: Amazon Textract
    Learners use the nine-part pattern on Amazon Textract document extraction. It covers text, forms, tables, queries, specialized APIs for expenses, IDs and lending documents, synchronous versus asynchronous processing, page-based pricing and limits.
19. Level 19: Amazon Comprehend
    Learners use the nine-part pattern on Amazon Comprehend natural language processing: entities, sentiment, PII detection, language detection, and custom classification and entity recognition. Topic modeling, event detection and prompt safety classification have been closed to new customers since 30 April 2026.
20. Level 20: Amazon Transcribe
    Learners use the nine-part pattern on Amazon Transcribe speech-to-text. It covers batch and streaming transcription, speaker diarization, custom vocabularies, PII redaction, Call Analytics and Medical, per-second pricing and the limits that bite.
21. Level 21: Polly and Translate: Speech Out, Languages Across
    Learners run Amazon Polly and Amazon Translate through all nine parts: what each one does, where it fits, the request and character limits that cause problems, per-character pricing and voice engine tiers, custom lexicons and terminology, and the smallest working synthesis and translation calls with clean-up afterwards.
22. Level 22: Kendra in Maintenance Mode: Run It or Migrate It
    Learners cover Amazon Kendra's index editions, connectors, query limits and hourly index pricing, then deal with its June 2026 move to maintenance mode and its closure to new customers from 30 July 2026 by planning a move to Bedrock Knowledge Bases.
23. Level 23: Personalize: Recommendations From Behaviour Data
    Learners cover Amazon Personalize's dataset groups, recipes, solutions, campaigns and batch jobs, including data minimums, training and campaign throughput costs, and a minimal recommender deployed and then torn down.
24. Level 24: Choosing and Evaluating Models
    Learners build a repeatable model selection process across Bedrock models, JumpStart models and custom models. The process combines task-specific evaluation sets, automatic and human or LLM-as-judge scoring, and latency and cost measured per successful task, not per token.
25. Level 25: Designing Retrieval-Augmented Generation on AWS
    Learners design a production RAG pipeline end to end: ingestion, chunking, embeddings, vector store choice, hybrid search, reranking, metadata and permission filtering, and retrieval evaluation. They compare managed Bedrock Knowledge Bases with a self-built pipeline.
26. Level 26: Responsible AI, Safety and Data Protection
    Learners apply safety and data protection controls across generative and predictive workloads: Bedrock Guardrails, PII handling, encryption with KMS, VPC endpoints, data residency and cross-Region inference, logging, and bias and explainability checks. Several SageMaker AI responsible AI features, including Clarify and A2I, entered maintenance mode in 2026.
27. Level 27: MLOps and LLMOps in Production
    Learners set up a delivery lifecycle for models and prompts: versioned data and artifacts, SageMaker Pipelines and Model Registry, CI/CD with approval gates, shadow and canary deployments, drift and quality monitoring, and prompt and evaluation versioning for generative apps.
28. Level 28: Worked Cost: A Generative AI Workload
    Learners cost a realistic RAG chatbot from end to end using dated example prices: input and output tokens, embeddings, vector store, guardrails and agent runtime. They then compare on-demand, batch, prompt caching and provisioned throughput, and model routing.
29. Level 29: Worked Cost: A Predictive ML Workload
    Learners cost a predictive workload such as fraud scoring or demand forecasting across training, tuning, storage and inference. They compare real-time, serverless, asynchronous and batch hosting, Spot training and Savings Plans, and weigh it against a ready-made API.
30. Level 30: Building the Weighted Selection Scorecard
    Learners turn the Level 2 workload profile into a weighted scorecard. The scorecard has criteria weights tied to business priorities, scoring rules backed by evidence, knockout criteria such as residency or maintenance-mode status, and a sensitivity check showing whether the winner holds up.
31. Level 31: Scoring AWS Against Another Cloud
    Learners score the chosen AWS design against an equivalent on another cloud, such as Azure AI Foundry or Google Vertex AI, using the same scorecard. Every claim must cite a dated source on model access, Regions, pricing, limits, compliance and lock-in.
32. Level 32: Final Challenge: Decision Workshop and Selection Record Exam
    In the final challenge, learners take a new workload scenario through the whole course: profile it, map it to candidate services, check retirement status, evaluate, design controls, cost it, score it, and compare it with another cloud. The result is a one-page selection record they can defend.

Access

The first 2 levels are free with a free account. Every level, the podcast edition and the AI tutor come with All Access at £4.99/month or any Creator plan — see pricing.

Start free →