AIP-C01: Build One GenAI Project That Covers the AWS Exam Blueprint

The AWS Certified Generative AI Developer - Professional exam (AIP-C01) has 5 domains, 20 task statements and well over 80 individual skills. You can study them as a list. Or you can do what the exam is really testing for: build a production-style generative AI application on AWS and make sure every part of the blueprint shows up somewhere in it.

This post maps the official AIP-C01 blueprint onto a single project you can build. Every exam figure comes from AWS's official AIP-C01 exam guide and the AWS certification page for the exam.

What is the AIP-C01 exam?

AIP-C01 is AWS's professional-level certification for developers who integrate foundation models into applications and business workflows, and run them in production on AWS.

AWS says the exam validates your ability to:

• design solutions with vector stores, Retrieval Augmented Generation (RAG) and knowledge bases
• integrate foundation models (FMs) into applications and workflows
• apply prompt engineering and prompt management
• implement agentic AI solutions
• optimise for cost, performance and business value
• implement security, governance and responsible AI
• troubleshoot, monitor and evaluate GenAI applications

Just as useful is what's out of scope: model development and training, advanced ML techniques, and data or feature engineering. This is an application developer's exam, not a data scientist's.

What are the AIP-C01 exam details?

75 questions (65 scored, 10 unscored) in 180 minutes, a passing score of 750 on a 100–1,000 scale, and a US$300 fee.

From the exam guide and the AWS certification page:

• Questions: 75 in total; 65 count towards your score and 10 are unscored and unidentified
• Question types: multiple choice and multiple response (you must select every correct option to get credit)
• Duration: 180 minutes
• Passing score: 750 (scaled, 100–1,000), with compensatory scoring, so you don't need to pass each domain separately
• Cost: US$300, plus any location-based fees
• Delivery: Pearson VUE test centre or online proctored
• Languages: English, Japanese, Korean and Simplified Chinese
• Prerequisites: none required

AWS recommends two or more years building production-grade applications on AWS or with open-source technologies, general AI/ML or data engineering experience, and one year of hands-on GenAI implementation. Unanswered questions are marked wrong and there's no guessing penalty, so never leave one blank.

How are the five AIP-C01 domains weighted?

Foundation model integration and data is the largest at 31%, followed by implementation (26%), safety and governance (20%), optimisation (12%) and testing (11%).

Domains 1 and 2 together are 57% of the scored content. In practice that means RAG, vector stores, prompt management, agents, deployment and API integration make up most of the exam.

What project covers the AIP-C01 blueprint?

An internal "ask our docs" assistant: a RAG application on Amazon Bedrock with guardrails, an agent that can call tools, and full monitoring and evaluation.

It's a common real-world pattern, and it naturally touches every domain. Build it in stages, and map each stage to the blueprint.

Stage 1: Retrieval and data (Domain 1, 31%)

Load a set of documents into Amazon S3 and build retrieval on top of them. This covers the largest domain:

• Task 1.4 and 1.5, vector stores and retrieval. Use Amazon Bedrock Knowledge Bases first, then try a second vector store, such as Amazon OpenSearch Service or Amazon Aurora with pgvector, so you can compare. Experiment with chunking strategies (fixed-size versus hierarchical), embedding models, hybrid search and reranking.
• Task 1.3, data pipelines. Add validation and clean-up before ingestion, and handle at least one non-text type, such as audio transcribed with Amazon Transcribe.
• Task 1.2, model selection and resilience. Make the model configurable without code changes (the guide mentions AWS AppConfig for this), and look at cross-Region inference for resilience.
• Task 1.6, prompts. Store prompts in Amazon Bedrock Prompt Management as versioned, parameterised templates rather than hard-coding them.

Stage 2: Agents and integration (Domain 2, 26%)

Give the assistant the ability to act, not just answer:

• Task 2.1, agentic AI. Add an agent that can call tools, such as looking up a ticket or creating one. The guide names Strands Agents, AWS Agent Squad and the Model Context Protocol (MCP), including MCP servers on AWS Lambda or Amazon ECS. Add safeguards: stopping conditions, timeouts and IAM boundaries.
• Task 2.4, API integration. Call models through the Bedrock APIs, add streaming responses for a real-time feel, and handle throttling with exponential backoff and fallbacks.
• Task 2.2 and 2.3, deployment and enterprise integration. Front the app with Amazon API Gateway, trigger ingestion from Amazon EventBridge events, and deploy it with a CI/CD pipeline.

If MCP is new to you, I covered what the Model Context Protocol Associate (MCPA) exam tests in an earlier post; it's a good primer on the protocol AIP-C01 keeps referring to.

Stage 3: Safety and governance (Domain 3, 20%)

Now make it safe to put in front of users:

• Task 3.1, input and output safety. Attach Amazon Bedrock Guardrails to filter harmful content and detect prompt attacks, and ground answers in your knowledge base to reduce hallucinations.
• Task 3.2, data privacy. Detect and mask PII (the guide mentions Amazon Comprehend and Amazon Macie), keep traffic private with VPC endpoints, and apply least-privilege IAM.
• Task 3.3 and 3.4, governance and responsible AI. Log decisions, track data sources, use AWS CloudTrail for audit, and show users the sources behind each answer.

A guardrail is a single parameter on the Bedrock Converse API, which makes it easy to wire in from the start:

import boto3 client = boto3.client("bedrock-runtime") response = client.converse( modelId="YOUR_MODEL_ID", messages=[{"role": "user", "content": [{"text": "Summarise our refund policy."}]}], guardrailConfig={"guardrailIdentifier": "YOUR_GUARDRAIL_ID", "guardrailVersion": "1"}, inferenceConfig={"maxTokens": 512, "temperature": 0.2}, ) print(response["output"]["message"]["content"][0]["text"]) 

Stage 4: Cost, performance and monitoring (Domain 4, 12%)

Production GenAI is as much about tokens and latency as about answers:

• Task 4.1, cost. Track token usage, trim context, cap response sizes, route simple questions to smaller models, and add caching (the guide lists semantic caching and prompt caching).
• Task 4.2, performance. Use streaming, parallel requests, and sensible temperature and top-p settings; tune retrieval for speed.
• Task 4.3, monitoring. Turn on Amazon Bedrock model invocation logging, build Amazon CloudWatch dashboards for token usage and errors, and trace calls with AWS X-Ray.

Stage 5: Evaluation and troubleshooting (Domain 5, 11%)

Finally, prove it works and learn how it fails:

• Task 5.1, evaluation. Create a small golden dataset of questions and expected answers, run Amazon Bedrock model evaluations, try LLM-as-a-judge scoring, and measure retrieval quality separately from answer quality.
• Task 5.2, troubleshooting. Deliberately break things: overflow the context window, use a chunk size that's far too large, change a prompt template and watch quality drop. Then diagnose each problem from logs and traces.

How long should you spend on each stage?

Roughly in proportion to the domain weights: about a third of your time on retrieval and data, a quarter on agents and integration, and the rest split across safety, optimisation and evaluation.

An eight-week plan for someone with the recommended experience:

1. Weeks 1–2: Stage 1, retrieval, vector stores, data pipelines and prompt management.
2. Weeks 3–4: Stage 2, agents, tools, MCP, APIs and deployment.
3. Week 5: Stage 3, guardrails, privacy and governance.
4. Week 6: Stages 4 and 5, cost, performance, monitoring and evaluation.
5. Week 7: Read the exam guide's skills line by line and tick off each one your project covers; fill the gaps.
6. Week 8: Timed practice at 75 questions in 180 minutes, reviewing every wrong answer against the guide.

If you're newer to AWS itself, AWS lists the AI Practitioner, Solutions Architect - Associate, Machine Learning Engineer - Associate and Data Engineer - Associate certifications as optional preparation. My 30-day AWS certification study plan is a lighter starting point for those.

What mistakes do AIP-C01 candidates make?

Studying model training, treating it as a pure Bedrock exam, and ignoring operations and evaluation.

• Studying out-of-scope ML. Training, feature engineering and advanced ML techniques are explicitly out of scope.
• Only learning Bedrock. Bedrock is central, but the guide also leans on Lambda, Step Functions, API Gateway, OpenSearch, SageMaker AI, CloudWatch, X-Ray and IAM.
• Skipping agents. Agentic AI, tools and MCP appear across Domains 1, 2 and 5.
• Treating safety as an afterthought. At 20%, safety, security and governance carry more weight than optimisation and testing combined.
• Not reading multiple-response questions carefully. You need every correct option for credit.

Is the AWS Certified Generative AI Developer - Professional worth it?

For developers already shipping GenAI features on AWS, yes. It certifies the production skills (RAG, agents, guardrails, monitoring and evaluation) that separate a demo from a real application.

It's a professional-level certification with no prerequisites, aimed squarely at people building with foundation models rather than training them. The US$300 fee and 180-minute exam reflect that level. If you're early in your AWS or GenAI journey, build experience first; the exam assumes a year of hands-on GenAI work.

AIP-C01 FAQ

How many questions are on the AIP-C01 exam?
75: 65 scored and 10 unscored.

How long is the exam?
180 minutes.

What is the passing score?
750 on a scaled score of 100–1,000.

How much does it cost?
US$300, plus any location-based fees.

Which domain carries the most weight?
Domain 1, Foundation Model Integration, Data Management, and Compliance, at 31%.

Is model training on the exam?
No. Model development and training are out of scope.

Are there prerequisites?
No required certifications. AWS recommends 2+ years building production applications and 1 year of hands-on GenAI work.

Can I take it online?
Yes, online proctored or at a Pearson VUE test centre.

Ready to start?

Pick the project, build it stage by stage, and check every skill in the exam guide against what you've built. By the time your assistant retrieves, acts, stays safe, reports its costs and survives your own fault testing, you'll have covered the AIP-C01 blueprint the way the exam expects.

For a full domain-by-domain breakdown with sample questions, this AWS AIP-C01 exam syllabus and study guide is a useful place to plan your preparation.

Exam details are from AWS's official AIP-C01 exam guide and certification page. Confirm current details before booking.