AWS Certified AI Practitioner - Concepts, Demos, Questions

Explore the fundamentals of ai and machine learning, clarify terminology, and outline learning techniques and deep learning processes. Preview ai life cycle and amazon ai and machine learning stack.

Explore key ai terminology—artificial intelligence, machine learning, deep learning, and generative ai—through real-world examples like voice assistants and the full machine learning process from problem to deployment.

Learn how training data quality drives model performance, differentiate labeled from unlabeled data and structured from unstructured data, and cover supervised, unsupervised, and reinforcement learning with spam detection.

Explore how a trained machine learning model uses brand new data to draw conclusions through batch and real-time inferencing, including real-world examples like stop-sign recognition.

Explore deep learning fundamentals, neural networks and generative AI, and learn the foundational models lifecycle from data selection to training, optimization, evaluation, and fine-tuning, with Amazon AI services.

Explore Amazon SageMaker and Bedrock to build, train, and deploy models with tools like Clarify and Data Wrangler, and access foundation models via a serverless API.

Explore real world AI use cases across retail, healthcare, and financial services, and discover computer vision, natural language processing, and ML techniques powering Amazon AI model support.

Explore AI use cases across industries, from computer vision in self-driving cars to natural language processing, document processing, and fraud detection in finance, and review supervised, unsupervised, and reinforcement learning.

Explore supervised learning with labeled data: classification for fraud detection and image classification, regression for weather or market forecasting, unsupervised clustering and dimensionality reduction, and reinforcement learning with AWS DeepRacer.

Explore gen AI capabilities—adaptability, personalization, scalability—and how models tailor content to tasks, while noting regulatory violations, social risk, data security and privacy concerns, toxicity, and hallucinations.

Discover Amazon Bedrock, a fully managed service that lets you choose foundation models, use playgrounds for text and images, and safeguard outputs while orchestrating data with RAG and knowledge bases.

Explore amazon bedrock foundation models, including titan embeddings and titan text g1 light express, and test via console and api calls for stability ai image and text prompts.

Explore how to implement safeguard guardrails in Amazon Bedrock to deny harmful topics and filter PII, then test and deploy guardrails for secure AI applications.

Explore Bedrock evaluation workflows by creating automatic or human evaluations, selecting a model such as Amazon Titan, choosing text summarization tasks, metrics, and S3-backed datasets with IAM access.

Learn prompts and prompt engineering to generate accurate ai outputs, including building blocks like instructions, context, examples, negative prompt technique, and zero-shot, few-shot, and chain-of-thought techniques.

Define a prompt as any input given to an AI model to produce a desired output. Use effective prompting to enhance capabilities and achieve high-quality, domain-specific results.

Explore the building blocks of prompts, including instruction, context, input and example, and output indicator, to guide artificial intelligence models toward precise, efficient results.

Use negative prompting to guide the model away from producing content and prevent hate speech, explicit language, bias, and include keywords such as worst quality, low quality, and low blurry.

Learn to craft prompts with clear context, output indicators, and examples, break complex tasks into simple steps, and explore prompt misuse risks like poisoning, prompt injection, and exposure.

Explore three prompt engineering techniques—zero-shot, few-shot, and chain-of-thought prompting—and optimization tips like larger foundation models and instruction tuning, with sentiment analysis examples.

Explore responsible AI practices, define responsible AI, discuss generative AI challenges, dimensions, and benefits, and review AWS services that mitigate risks for exam readiness.

Explore responsible ai principles that ensure transparency and trustworthiness while mitigating bias across design, data, and interaction. Learn mitigation strategies—diverse data, explainability, diverse stakeholders, and auditing.

Explore gen AI challenges such as hallucination, toxicity, intellectual property concerns, cheating, and disruption of work, with exam emphasis on understanding these risks.

Explore the dimensions of responsible AI, including fairness, explainability, privacy and security, transparency, veracity and robustness, governance, and controllability to ensure unbiased, secure, and human-aligned AI.

Discover the benefits of responsible AI, including ethical, transparent, and accountable use, regulatory compliance, explainable and auditable applications, increased trust, competitive advantage, and risk mitigation against threats like prompt injection.

Compare models in Bedrock and SageMaker to evaluate latency and accuracy, then use SageMaker Clarify to assess metrics like accuracy, robustness, and toxicity.

Explore Amazon Bedrock guardrails to filter inappropriate content and remove PII, and use SageMaker Clarify and Data Wrangler to detect bias and balance data.

Explore transparency and governance through AI service cards detailing use cases, limitations, responsible AI design, deployment guidance, and monitor with SageMaker model monitor and Amazon Augmented AI for human review.

Explore model optimization by transforming text to vector embeddings, using gen AI agents, and evaluating with human feedback and benchmarking, then compare tuning methods and metrics like rogue and blue.

Explore how foundation models use vector embeddings to convert text and media into vectors stored in vector stores for fast, scalable enterprise search with AWS OpenSearch, RDS Postgres, and Kendra.

Explore generative AI agents, software entities that use models to simulate human responses, bridging back-end systems, launching actions, and integrating feedback to speed up AI workflows.

Evaluate GenAI results using a combined approach that blends human feedback on user experience, contextual appropriateness, creativity, and flexibility with benchmark datasets that measure accuracy, speed, efficiency, and scalability.

Fine-tune foundational models to boost output specificity, accuracy, and efficiency with domain-specific, diverse data, covering instruction tuning, reinforcement learning from human feedback, domain adaptation, transfer learning, and continuous pre-training.

Explore model evaluation metrics such as ROUGE (ROUGE-N and ROUGE-L) for summarization and translation quality, BLUE for translation, and BIRD score for semantic similarity.

Explore the concepts of security, governance, and compliance, define each term, review AWS compliance services, data governance, and AI security along with data and model lineage.

Review AWS config for resource configurations and audit entries across EC2 and RDS, while AWS Inspector scans for risks and CloudTrail logs API calls for compliance.

Explore data governance for AI apps, covering data life cycle, residency, logging, disposal, encryption in transit, monitoring, and retention. Address governance strategies, transparency, bias checks, and responsible AI practices.

Track data and model lineage by capturing origin, transformations, and movement; catalog datasets and models, then document model cards with use cases, performance, and limitations in Amazon SageMaker.