Subject
AWS AI Practitioner
Twelve articles following the AWS Certified AI Practitioner (AIF-C01) exam guide in order: what AI, machine learning and generative AI are, which AWS service does which job, how applications are built on foundation models with Bedrock, retrieval and prompts, how models are customised and judged, and how to keep all of it responsible, secure and compliant. Written to be read once in order and revised from the maps and question sets.
Articles
-
1 · The AIF-C01 exam and the AWS AI stack: what is tested and where each service sits
The exam's format, scoring and five weighted domains, the four layers of the AWS AI stack, the scenario keywords that decide answers, and how the series follows the blueprint.
-
2 · AI, machine learning and deep learning: the vocabulary everything else assumes
AI against ML against deep learning, neural networks in plain words, supervised, unsupervised and reinforcement learning, regression, classification and clustering, batch and real-time inference, data types, fit, bias and the algorithm families by name.
-
3 · When machine learning fits, and which AWS AI service does what
When a model is the wrong tool, matching techniques to problems, and the full catalogue of AWS managed AI services from Rekognition to Personalize with the keyword that points at each.
-
4 · The machine learning lifecycle on SageMaker: from raw data to a monitored endpoint
Every stage of the ML pipeline with the SageMaker feature for it, the four ways to deploy a model, MLOps, and the model and business metrics the exam asks about.
-
5 · Generative AI fundamentals: tokens, embeddings, transformers and foundation models
How text becomes tokens and vectors, what a transformer does, what makes a model a foundation model, diffusion and multimodal models, the use cases, and the foundation model lifecycle.
-
6 · What generative AI can and cannot do, and the AWS stack built for it
Hallucination, nondeterminism and the other limits, how to pick a model, the business metrics, and Bedrock, SageMaker JumpStart, PartyRock and Amazon Q with their pricing trade-offs.
-
7 · Amazon Bedrock: one API for foundation models, knowledge bases, agents and guardrails
The model catalogue, inference parameters, Knowledge Bases, Agents, Guardrails, fine-tuning and continued pre-training, model evaluation, pricing and security, in one reference.
-
8 · Designing applications on foundation models: model choice, RAG, vector stores and agents
Selection criteria, retrieval augmented generation end to end, the AWS vector stores and when each fits, the cost ladder from prompting to pre-training, and what agents add.
-
9 · Prompt engineering: getting the answer you want, and the attacks that get answers you do not
The anatomy of a prompt, zero-shot to few-shot and chain-of-thought, templates and best practices, and the risks: exposure, poisoning, injection and jailbreaking, with their defences.
-
10 · Customising and evaluating foundation models: fine-tuning, RLHF, ROUGE, BLEU and benchmarks
Pre-training, fine-tuning methods and RLHF, preparing data, and judging the result with human evaluation, benchmark datasets, ROUGE, BLEU and BERTScore against business objectives.
-
11 · Responsible AI: bias, fairness, explainability and the AWS tools that measure them
The dimensions of responsible AI, the kinds of bias and where they enter, SageMaker Clarify, Model Monitor, Augmented AI, Guardrails and Model Cards, and the trade-off between safety and transparency.
-
12 · Security, compliance and governance for AI solutions on AWS
IAM, encryption, Macie and PrivateLink, the shared responsibility model, data lineage and privacy techniques, the standards, and Config, Inspector, Audit Manager, Artifact, CloudTrail and Trusted Advisor told apart.