AWS Certified Generative AI Developer Professional (AIP-C01)

Explore production-ready gen AI applications for wind turbines, guiding field technicians with an equipment SME assistant built on Bedrock, driven by 15 architectural decisions.

Explore the gen-ai production architecture through 15 key decisions, from Bedrock or SageMaker Jumpstart and foundation model selection to guardrails, orchestration, cost optimization, and security.

Explore Amazon Bedrock, a fully managed, serverless genai service offering base models from Amazon and third parties via API, with model IDs, prompts, and inference parameters.

Explore the Amazon Bedrock reference architecture: Bedrock service account, model provider escrow account, and runtime inference routing to the right model endpoint. See requests flow via SDK, console, or CLI.

Explore the Amazon Bedrock console and its model catalog, including serverless vs marketplace deployments, API keys, testing, tokenizer, watermark detection, cross region inference, and prompt routers.

Explore how Amazon Bedrock enables drag-and-drop low-code workflows with flows, knowledge bases, guardrails, prompt management, and data automation to build generative AI applications.

Access and deploy foundation models with Amazon SageMaker JumpStart, fine-tune them, and evaluate performance within your own VPC, using open-source and proprietary models.

Assess foundation model modality, task complexity, and context window, then evaluate guardrails, accuracy, latency, cost per 1000 tokens, regional availability, and language support to select the smallest, safest, cost-efficient model.

Explore foundation model selection in AWS Bedrock by comparing modalities, context windows, parameters, costs, and quotas to choose the right model for your use case.

Explore how foundation model inference parameters shape output, controlling randomness and diversity with temperature, top k, and top p, and length with max tokens and stop sequence.

Tune inference parameters on Amazon Bedrock by adjusting max output tokens, stop sequences, and latency. Explore temperature, top k, and top P to shape creativity and variety.

Configure Amazon Bedrock guardrails to enforce organization policies, block investment advice, and filter prompts with content filters, denied topics, and contextual grounding checks.

Explore how Amazon Bedrock Automated Reasoning checks validate LLM outputs against a domain knowledge policy to reduce hallucinations and enforce guardrails in applications like home insurance claims.

Use prompt engineering to optimize input to foundation models, improving completions and inference; explore zero-shot, one-shot, and few-shot prompting, plus chain-of-thought and react reasoning with retrieval augmented generation.

Discover how Amazon Bedrock prompt management creates, evaluates, versions, and shares prompt templates and system prompts to improve responses from foundation models, using a wind turbine equipment SME example.

Explore Bedrock prompt management to create, test, and reuse prompts across GenAI applications, configure system and user prompts, and refine inference parameters with Nova Pro models for optimized responses.

Explore inference optimization for generative AI by examining Amazon Bedrock service tiers (reserve, priority, standard, flex) and how capacity planning and token-based pricing shape performance.

Enable cross-region inference with inference profiles to route requests between Bedrock regions, boosting throughput. Choose global or geographic options for data residency, and apply latency settings to speed responses.

Demonstrates inference optimization with a cross region inference profile for Bedrock models. Shows how to implement the demo using Boto3 code, a Lambda function, optimized latency, and a guardrail.

Explore Bedrock pricing models for generative AI: on-demand, provisioned throughput, and batch inferencing, with token-based costs, throughput guarantees, and overnight batch processing in S3.

Understand Amazon Bedrock pricing, including badge inference at 50% lower than on-demand, on-demand per 1000 input/output tokens, and provisioned throughput pricing for model units with monthly commitments.

Explore how Amazon Bedrock prompt caching cuts latency and input token cost by reusing the same document and system prompt for repeated inquiries, such as turbine logs for machine IDs.

Explore a hands-on demo of Amazon Bedrock prompt caching using a lambda function, highlighting static system prompts, dynamic user prompts, and machine id-based token and latency reductions.

Explore Amazon Bedrock Intelligent Prompt Routing that dynamically routes requests among foundation models from family, like Nova Lite and Nova Pro, to optimize cost and quality with thresholds on quality differences.

Configure the Amazon Bedrock intelligent prompt router to route between Nova Pro and Nova Lite using a 10% threshold, with Nova Pro as fallback and the ARN guiding programmatic requests.

Evaluate the seventh architectural decision by examining EC2 as a long-running compute option for JNI workloads, including orchestration, prompt assembly, training and hosting, and rag pipelines.

Harness AWS Lambda as a fully managed, event-driven serverless platform for Gen-AI orchestration in Python or Java, with pay as you use pricing and a 15-minute limit.

Explore Amazon ECS as a fully managed, scalable compute layer for GenAI workloads, comparing Fargate and ECS on EC2, with pricing, scaling, and long-running inference use cases.

Explore how to choose the Gen-AI API layer with Amazon API Gateway, comparing REST API, HTTP API, and WebSocket API for secure, scalable prompt routing in a Bedrock Lambda workflow.

Explore AWS AppSync as the GenAI API layer for real-time chat and live dashboards with GraphQL, fetching data from multiple sources in one call, with Lambda and Bedrock integration.

Discover how the application load balancer, a layer-7 router, distributes GenAI API requests across EC2, ECS, EKS, or AWS Lambda for high availability and low-latency inferences.

Monitor a GenAI application on Amazon Bedrock using CloudWatch metrics, logs, and log insights. Track invocation latency, input/output token counts, and errors, set alarms for thresholds, and analyze guardrails impact.

Enable model invocation logging in Amazon Bedrock and route logs to CloudWatch. Then monitor metrics, token counts, and latency with CloudWatch logs insights and dashboards.

Explore evaluating foundation models with Amazon Bedrock model evaluation, including creating a prompt data set and choosing evaluation methods such as programmatic llm as a judge or human evaluation.

Set up and run a hands-on model evaluation in Amazon Bedrock, using an LLM as a judge and selecting quality and responsible AI metrics.

Explore model distillation as a Bedrock customization, transferring knowledge from a large teacher model to a smaller student model using prompts to achieve near large model accuracy with lower latency.

Perform hands-on model customization on Amazon Bedrock, using distillation to train a student model from a teacher, with synthetic data, an S3 bucket, and supervised or reinforcement fine tuning.

Apply supervised fine-tuning with labeled prompt and response pairs to tailor outputs for tasks like summarization and Q&A, using Bedrock and SageMaker to train a fine-tuned LLM.

Continue pre-training with unlabeled data to boost domain knowledge for medical summarization. Train on proprietary medical journals to tailor a foundation model.

Explore fine-tuning approaches, focusing on LoRa, a parameter-efficient method that freezes base weights and trains a small set of added parameters for healthcare terminology domain adaptation and multi-language adaptation.

Explore architectural decision points 12–15, covering security and data protection, enterprise data management, gen ai and ops and governance, and orchestration with AWS Lambda.

Explore retrieval augmented generation (rag) and ten architectural considerations for rag apps, framed by an e-learning q&a app referencing PDFs in a S3 bucket and whitelisting bedrock and approved models.

Learn why rag, or retrieval augmented generation, is needed to overcome foundation model limitations, such as missing enterprise data and outdated knowledge, by fetching information from internal sources and documents.

Explore vectors, embeddings, and semantic search to perform similarity-based retrieval from vector stores, using chunking and embedding models across Amazon Bedrock to answer questions.

Explore ten architecture decisions to build production-ready rag apps for e-learning, from ingestion pipeline and data sources to vector databases, embedding models, re-ranker, llm selection, and observability.

Explore Amazon Bedrock Knowledge Basis, a fully managed rack connecting private data to foundation models. Learn options for data sources, chunking, embeddings, vector databases, and Retrieve API and Retrieve-and-Generate API.

Build a Bedrock knowledge base from an S3 bucket and configure the data source, parsing, chunking, and embeddings. Test retrieval-only and retrieval-and-generation modes with vector stores, citations, and guardrails.

Learn the first rag architecture decision: data sources and data types, from structured data from Redshift to unstructured PDFs and multimodal content, with bedrock knowledge base.

Configure rag data sources in AWS Bedrock by building structured and unstructured knowledge bases and wiring metadata from Redshift or Glue to connect sources like S3, Confluence, and Salesforce.

Compare chunking strategies for rag-based apps to balance retrieval precision, processing cost, and accuracy, from no chunking to fixed, hierarchical, semantic, and cementing approaches, with practical use cases.

Choose the embedding model by input modality (text or text plus image) and output dimension, balancing quality, cost, latency, and multilingual needs.

Compare AWS vector DB options for a rack-based app: Aurora PostgreSQL with PG Vector, Neptune Analytics, OpenSearch, Pinecone, and Redis for RAG workloads.

Explore retrieval optimization for rag based apps using re-ranker models and metadata filtering, demonstrating how top-relevance reordering improves accuracy, reduces hallucinations, and lowers token usage.

Explore metadata filtering as a retrieval optimization technique, using metadata tags to restrict vector searches by attributes like country and department, reducing cost and improving accuracy.

Select the right large language model for generation in rag workloads using the ten key parameters from foundation model selection, applied to our rag based e-learning app.

Explore RAG observability by using ingestion and resource logs in CloudWatch to monitor knowledge base data loading, embedding generation, and runtime API usage for reliable responses.

Explore hands-on RAG monitoring by wiring a Bedrock knowledge base to CloudWatch logs, syncing data through chunking and vector embeddings into the vector store.

Explore rag evaluation with bedrock to assess models and knowledge bases, including outside-bedrock sources, using retrieval and generate metrics, context coverage, relevance, and responsible ai checks.

Explore RAG security and guardrails for responsible AI, and learn about architectural decision points related to these topics. The course provides deep dives in dedicated sections for these subjects.

Amazon Q business is a managed generative AI powered assistant that boosts productivity by chatting with enterprise data, summarizing documents, and generating draft content with plugins like Jira and ServiceNow.

Explore how AI agents address complex tasks and the limitations of LLMs in leave balance checks, weather, and bookings, with retrieval augmented generation and agentic approaches.

Discover how AI agents, driven by large language models, decompose tasks, use tools, remember interactions, and enforce guardrails to autonomously plan and execute complex travel scenarios.

Discover how AI agents plan with chain of thought, tree of thought, and react prompting, using a few short prompts, observations, and dynamic re-planning for tasks.

Watch an AI agent check an employee's leave balance and, if eligible, book a hotel via an API, illustrating planning, tracing, and decision making.

Explore how AI agents use tool calling to connect external and internal systems, using weather APIs, hotel bookings, databases, a code interpreter, and rag base solutions.

Explore how a micro cool ai agent uses three tools—the bedrock knowledge basis, the leave balance API, and the hotel booking API—to answer HR queries, check balances, and book rooms.

Explore how AI agents use short term memory for session context and long term memory to summarize past interactions, enabling personalization with DynamoDB-backed context.

Discover Amazon Bedrock agents, a fully managed service that automates multistep tasks by linking with systems and data, using react orchestration, planning and reasoning, and action groups and knowledge bases.

Explore the Amazon Bedrock Agents console by creating a flight booking agent and enabling multi-agent collaboration, and configure models, action groups, memory, and orchestration.

Discover how Amazon Bedrock agents work by augmenting user input with a prompt store, tools, and session history; perform task planning, tool execution, and iterative observations to deliver final responses.

Explore strands agent, an open source SDK using a model driven, large language model reasoning approach to build AI agents with code, model agnostic and integrated with AWS Bedrock.

Explore AgentSCOD, an open source framework that orchestrates multiple AI agents through a classifier, routing requests like flight, hotel, and itinerary tasks, and storing histories in DynamoDB.

Explore the difference between generative AI and agentic AI, both powered by language models, contrasting content generation with autonomous task execution, and show how memory and data access drive approaches.

Explore model context protocol, or MCP, and how it standardizes how LLM applications access internal and external data sources through a host, client, and server with tools, resources, and prompts.

Demonstrates installing the AWS CloudFormation MCP server with the QCLI client, building CloudFormation templates via natural language, and deploying resources like S3, Lambda, and API Gateway.

Explore how Amazon Bedrock Agent Core enables moving agentic AI apps from proof-of-concept to production using enterprise-ready capabilities, demonstrated through a dream vacation planner workflow with research and summarizer agents.

Learn how the AWS Bedrock Agent Core Runtime deploys agentic AI code as a container, exposes endpoints, manages identity and memory, and enables gateway, browser, and interpreter tools with observability.

Explore the Gen AI orchestration layer and why coordinating steps, retries, and human-in-the-loop decisions matters for enterprise AI. See how a travel-booking workflow uses Lambda and Step Functions.

Analyze how AWS step functions orchestrate generative ai workflows with state machines and tasks, featuring a card recommendation use case with pii detection, tokenization, and audit history.

Compare standard and express AWS Step Functions workflows, highlighting long-running tasks like e-commerce orders and ETL, with standard offering audit logs and exactly-once execution, and express at-least-once guarantees.

Discover Amazon Bedrock Flows, a visual drag-and-drop workflow builder for generative AI that uses node-based orchestration to connect Bedrock foundation models, knowledge bases, and AWS services for prototyping and deployment.

Demonstrate building and testing a Bedrock flow by dragging nodes, linking prompt agents and knowledge bases, and configuring Nova Pro prompts to retrieve or generate responses.

Apply a mental model to select AWS GenAI services: lambda for simple orchestration, bedrock knowledge base for retrieval, bedrock agents for autonomous reasoning, and step functions for deterministic workflows.

Explore Bedrock Data Automation, which extracts fields from unstructured multi-modal content and formats results to your backend schema using standard and custom outputs, blueprints, and projects.

Develop data foundations for generative AI on AWS by collecting, cleansing, and preparing data, then train, evaluate, and deploy models using SageMaker Data Wrangler, Feature Store, Clarify, and Model Monitor.

Explore sage maker data wrangler for data collection from S3, Athena, sources. Then perform exploratory data analysis, apply pre-built or custom transformations, and export to S3 or feature store.

Learn how the Amazon SageMaker feature store ingests streaming or batch data, catalogs features in the feature catalog and feature groups, and supports online real-time inference and offline training.

Explore AWS's eight guiding principles of responsible AI—veracity and robustness, safety, privacy and security, explainability, fairness, transparency, controllability, and governance—and see real-world examples and tools to build safe, ethical AI.

Explore how Amazon SageMaker Clarify detects bias in pre-training and post-training data and models, explains predictions, and highlights influential features, while SageMaker Model Monitor tracks data quality and production performance.

Explore how Amazon SageMaker model monitor safeguards production inference by tracking data quality and model performance, and how governance tools—model cards, model registry, and role manager—support responsible AI.

Explore SageMaker pipelines for end-to-end ML workflows at scale, featuring data processing, training, tuning, evaluation, and deployment, plus SageMaker ground truth for RLHF with human labeling.

Explore Amazon Augmented AI, a service that routes high and low confidence predictions to human reviewers and enables workflows for insurance, forms, and content moderation with Textract and Rekognition.

Map AWS AI services into machine learning, neural networks and deep learning, computer vision with amazon recognition service, and natural language processing with amazon transcribe and amazon poly.

Explore Amazon Rekognition's computer vision capabilities to detect, classify, blur, or delete inappropriate content in images; verify identities online, and analyze and tag live or recorded videos.

Explore Amazon Rekognition's label detection, image properties, and facial analysis in a hands-on session, plus a real-world workflow tagging and searching media assets with S3, SQS, Lambda, and Elastic Search.

Explore Amazon Lex, an AWS AI service, for building voice and text interfaces. Learn intents, utterances, slots, confirmation, and fulfillment through self-service and enterprise bot use cases.

Build a flight booking bot with Amazon Lex and Gen AI (Claude v2), defining intents, slots, and utterances, then test and integrate with Lambda for booking confirmation.

Explore how Amazon Transcribe converts speech to text and enables real-time and post-call analytics for medical transcription, call center analytics, and toxic content detection.

Explore Amazon Translate, an AWS service that translates across 75 languages in real time or in batches via S3, with automatic language detection and profanity masking.

Explore Amazon Translate, enabling fast batch and real-time translations with auto-detect, S3 input/output, and multi-format support, plus customization like formality, profanity masking, and custom terminology across 75 languages.

Explore how Amazon Comprehend analyzes text from emails, social media, and documents to detect PII, perform sentiment analysis, recognize entities, and extract key phrases.

Amazon Polly turns text into lifelike speech using deep learning, converting text to audio. Use Polly for e-learning notes, content feeds, and call center applications.

Macie detects PI data in S3 and triggers Lambda redaction via EventBridge, while AWS PrivateLink enables private connectivity through gateway and interface endpoints, including on-premises access via Direct Connect.

Learn how AWS Identity and Access Management authenticates users and authorizes actions, using IAM users, groups, roles, and policies to govern access to services like S3, EC2, and DynamoDB.

Demonstrate configuring iam roles and policies to enable a lambda function to access amazon bedrock's coherent command light model, including creating a demo role and attaching bedrock full access policy.

Explore how AWS Config monitors resource changes and enforces policies, how AWS CloudTrail tracks who changed what, and how Amazon CloudWatch collects metrics, logs, and alarms—governance, regulation, and compliance.

Explore Amazon Inspector's role in compute workloads by scanning for vulnerabilities and unintended network exposure, and Trusted Advisor's six categories—cost optimization, security, resilience, performance, operational excellence, service limits—for remediation.

Explore aws artifact for on-demand compliance documents and aws audit manager for mapping data from aws services to industry frameworks, automating audit evidence and assessments.

Understand governance and compliance frameworks guiding AWS cloud use. Learn EU AI Act risk levels and the required conformity steps, plus the Algorithmic Accountability Act.

Understand the five scopes of the generative ai security scoping matrix—from customer apps with public services to self-trained models—and the changing governance, compliance, and ownership across scope one to five.