AWS Certified Generative AI Developer – Professional (AIP-C01)
Exam Notes & Practice Tests

A SaaS platform is building a multi-tenant GenAI-powered documentation assistant that allows enterprise customers to upload internal manuals and policies. The assistant performs semantic search using embeddings stored in a vector database and must guarantee strict tenant isolation, encryption at rest, secure access controls, and predictable performance as the number of customers grows into the thousands. Which architecture best meets these requirements?

A pharmaceutical research organization maintains a large corpus of drug safety documentation that is updated frequently. Minor editorial changes should trigger incremental embedding updates, while urgent safety corrections must be reflected immediately. Additionally, the organization requires a full rebuild of the vector index every quarter to eliminate drift and stale embeddings. Which maintenance strategy should the team implement?

A financial services firm has collected years of emails, chat transcripts, and support tickets across multiple storage systems. Before using this data for retrieval-augmented generation and model fine-tuning, the security team requires automated discovery of sensitive information such as PII, financial identifiers, and regulated data. Leadership also wants to ensure that only approved datasets flow into GenAI pipelines. What is the best approach?

A retail company operates a GenAI-powered customer service assistant used across web and mobile apps. Traffic increases dramatically during holiday sales and promotional campaigns, leading to throttling and inconsistent response times. Finance teams want predictable monthly costs, while product teams require consistent low-latency responses even during peak demand. How should the deployment be configured to balance performance and cost?

An insurance company is building a Claims Risk Assistant that evaluates structured inputs such as claim amounts, event types, historical risk scores, and categorical flags to generate short, structured risk summaries. The system must respond in under 400 milliseconds, scale to millions of requests per day, minimize inference cost, and strictly enforce a JSON output schema for downstream analytics and compliance reporting. Which solution best meets these requirements?

A government agency generates multilingual public notices using a foundation model. Leadership wants to ensure that tone and examples remain fair across demographic groups and that bias regressions are detected automatically over time. How should you implement continuous fairness evaluation?

A claims processing assistant routes insurance claims to different specialized models based on language, length, and claim type. The company wants routing logic that adapts over time using performance metrics such as cost, latency, and human review scores. Which two approaches best support intelligent, metric driven routing? (Select two)

A SaaS platform uses multiple foundation models and wants to shift traffic between them for different customer tiers without redeploying application code. Product managers also require gradual rollouts with monitoring and rollback. Which configuration driven design is MOST appropriate?

A technical support chatbot retrieves documents from product manuals, FAQs, and troubleshooting guides. Vector search improves recall, but irrelevant documents sometimes rank too high when product identifiers or model numbers matter. Which retrieval architecture best improves ranking precision for FM augmentation?

A vacation planning assistant handles complex user requests that include destination discovery, lodging preferences, and itinerary creation. Passing the raw query directly to retrieval produces noisy results and misses constraints. How should you structure query handling to improve retrieval quality?

A global telecommunications provider maintains a graph representing customers, network nodes, service plans, outages, and risk indicators. The company is now attaching support tickets, outage reports, and engineering notes as text associated with nodes and edges. The GenAI assistant must answer questions that depend on both network relationships and semantic similarity. What is the most effective approach to design a graph-aware retrieval mechanism for foundation model augmentation?

A healthcare payments company is deploying an internal GenAI assistant on Amazon Bedrock to analyze highly sensitive claims narratives. Compliance requirements state that all Bedrock API traffic must remain within the organization’s private AWS network and that only approved application subnets may access Bedrock, with zero public internet exposure. The solution must scale across Availability Zones. How should the network architecture be designed?

A large online learning platform runs an AI-powered teaching assistant used by students worldwide. Security teams have observed repeated attempts to bypass safety policies using prompt injection, role manipulation, and jailbreak-style inputs. The platform needs a real-time detection system that evaluates every incoming prompt, identifies adversarial patterns, and blocks high-risk requests before they reach the model. Which option best meets this requirement?

A research institution is deploying a very large language model—hundreds of gigabytes in size—for private inference within AWS. During deployment, containers fail health checks because model downloads and GPU initialization exceed default startup timeouts. Operations teams need predictable readiness behavior and must prevent rollback during long initialization phases. How should the hosting environment be configured?

A multi-tenant enterprise AI platform supports use cases such as compliance analysis, customer support automation, technical troubleshooting, and content generation. Each use case has different cost, latency, and accuracy requirements. The platform team wants to automatically route incoming prompts to the most appropriate foundation model without exposing model choices to end users. Which routing strategy should be implemented?

A global fintech company operates a GenAI-powered transaction insight service that currently runs in a single AWS Region. New resilience and regulatory mandates require the service to continue operating during a full regional outage, while ensuring that traffic is routed only to Regions approved for specific customer jurisdictions. The architecture must support automated health checks and seamless traffic redirection with minimal downtime. How should you architect the multi-Region integration and routing strategy to meet these requirements?

A risk management firm is building a retrieval-augmented assistant using Amazon Bedrock to analyze thousands of regulatory guidelines and compliance manuals stored as PDFs in Amazon S3. These documents contain nested sections, cross-references, tables, and footnotes. Early tests reveal that retrieval often omits key definitions because document chunking loses hierarchical context. What is the most effective way to redesign document preprocessing and chunking?

An enterprise search assistant continues to surface outdated operational guidance even after new documentation has replaced older content. Investigation shows that part of the corpus was re-embedded using an updated embedding model, while legacy vectors remain unchanged, leading to inconsistent retrieval results. How should embedding drift be identified and corrected to ensure accurate and consistent search behavior?

A centralized AI platform account hosts Amazon Bedrock for all foundation model inference across an organization. Individual departments maintain their own vector stores using Amazon OpenSearch Serverless or Aurora PostgreSQL with pgvector in separate AWS accounts. The AI platform must retrieve vectors directly from these remote stores without duplicating data or weakening security boundaries. What is the most effective architecture to support secure, low-overhead cross-account retrieval?

A public benefits agency uses a GenAI assistant to draft eligibility summaries for citizens. Auditors are concerned about gradual bias drift affecting different demographic groups over time. The agency requires continuous monitoring of fairness metrics, automatic alerting when thresholds are exceeded, and immediate remediation actions without manual intervention. How should this monitoring and response solution be designed?

A developer platform provides a GenAI-powered content moderation API that is consumed by hundreds of third-party applications. Requests vary by customer tier, and some clients send malformed JSON payloads or extremely large prompts that exceed model limits. During peak events, sudden request bursts can overwhelm the inference backend. The platform must validate request structure and enforce per-customer rate limits before any request reaches Amazon Bedrock. Which options should be implemented? (Select two)

A regulated insurance provider is building multiple GenAI-driven compliance assistants across development, testing, and production environments. Each assistant requires controlled prompt updates, approval workflows, rollback support, and full auditability of who approved each change and which configuration is active per environment. The organization wants to minimize custom infrastructure and operational overhead. Which solution best meets these requirements?

A multinational payments company must deploy GenAI inference capabilities close to sensitive customer data sources to satisfy national data residency laws. At the same time, it needs low-latency inference for users in large metropolitan areas and secure integration between on-premises systems, edge locations, and AWS Regions. Which architecture best satisfies these requirements?

A scientific research organization uses a GenAI assistant to answer questions over lengthy engineering specifications that routinely exceed the model’s context window. Passing entire document sections into prompts increases cost and latency and often truncates key clauses. The team wants to redesign prompt construction so only the most relevant content is included, while preserving accuracy. What approach should be used?

A global publishing company is building a GenAI-powered search assistant backed by more than 100 million archived articles stored in Amazon S3. The system must generate embeddings using Amazon Titan Text Embeddings for historical content and continuously embed tens of thousands of new articles per day. The solution must scale efficiently, tolerate failures, and avoid throttling during peak ingestion. Which approaches should be used? (Select two)

A large e-commerce company operates a GenAI-powered shopping assistant and wants to proactively test its safety mechanisms against new prompt-injection techniques. The security team plans to regularly inject synthetic role-override prompts, policy bypass attempts, and manipulation instructions to ensure the assistant consistently enforces safety policies. These tests must run automatically on a schedule, validate outputs for violations, and trigger alerts and reports when failures are detected. Which of the following options would you recommend for this scenario?

A telecommunications provider wants to summarize customer complaint tickets using a foundation model. Many tickets include sensitive data such as phone numbers, billing identifiers, and residential addresses. The company requires that sensitive data be detected and transformed before model invocation, and that all redaction actions be auditable for regulatory review. Which of the following represents the best solution?

A fintech startup is evaluating multiple large language models to power its contract analysis platform. Before allowing any model into production, the team must run standardized evaluations to compare reasoning quality, factual accuracy, and robustness under controlled conditions. The process must be repeatable, automated, and produce quantifiable scores. Which evaluation approach is most suitable?

A hospital network is deploying a GenAI assistant to generate summaries from patient encounter notes for multiple roles, including physicians, nurses, and care coordinators. The system must redact PHI before AI processing, support persona-specific summaries, and stream responses back to clinical applications with minimal latency. The organization prefers a managed orchestration approach that can grow as new workflows and personas are introduced. Which solution should the GenAI architect choose?

A multinational bank wants to generate monthly AI governance reports covering all GenAI workloads across Regions and departments. The reports must automatically summarize metrics such as bias incidents, drift detection, lineage completeness, guardrail enforcement, and remediation actions, without manual data collection from individual teams. How should this reporting architecture be designed?

A financial planning startup wants to build a customer-facing GenAI assistant that can remember user preferences across multiple chat sessions, such as risk appetite, preferred investment horizon, and previously approved recommendations. Which solution should you implement?

A global retailer wants to build a GenAI customer support assistant that responds to users in real time through a web and mobile app. The assistant must answer natural language questions, call internal APIs to fetch order status or refund eligibility, and sometimes trigger workflows to update CRM records. The engineering team wants a managed orchestration approach that can maintain conversation context and return responses within a few seconds. Which orchestration approach should the solutions architect recommend?

A customer support team is launching a new GenAI chat assistant that must stream responses to agents in real time across both web and mobile dashboards. The product team requires a typing indicator, smooth progressive display of generated text, and a responsive experience even when users are on unstable or intermittent mobile networks. How should you design the streaming interaction pattern and connection handling?

An HR SaaS provider is developing multiple GenAI capabilities on Amazon Bedrock, including job description generation, interview question drafting, and performance review summarization. Leadership wants centralized prompt governance with approvals, versioning, and auditability. How should you implement prompt management?

An e-commerce company is evaluating a new foundation model on Amazon Bedrock to improve product recommendation quality. The team wants to run an online A/B test, collect per-model metrics, and optionally support ensemble responses. How should you design the routing pattern?

A content moderation team performs nightly processing of millions of user posts using a large language model to detect policy violations. The workload is heavy but not user facing, and completing it within several hours is acceptable. Their current real time endpoint scales sharply during the batch window, causing high costs and occasional throttling. The team wants a design that can absorb very high volumes of requests, smooth out load, and better align infrastructure usage with the relaxed latency requirements of this batch job. How should you redesign the deployment to handle this large batch workload more efficiently?

A financial services company is building a retrieval augmented generation (RAG) system on Amazon Bedrock. New policy documents arrive in a central S3 bucket multiple times per day. Each document must pass through a pipeline that performs text extraction, PII redaction, chunking, embedding generation, and index updates in a vector database. The development team wants a repeatable, event driven orchestration pattern that separates ingestion from query time and can be versioned as the pipeline evolves. Which orchestration strategy would you use to implement and manage this GenAI ingestion workflow?

A financial services company uses a mix of third-party and custom models for different GenAI workloads, such as summarization, document classification, and code generation. Some workloads can use managed foundation models through Amazon Bedrock, while others must run on custom fine-tuned models hosted on SageMaker AI for compliance reasons. The company wants a single API layer that can route each request to the most appropriate model based on task type, latency requirements, and cost profiles. How should you design a hybrid deployment strategy and orchestration layer that integrates Amazon Bedrock and SageMaker AI endpoints in a secure and observable way?

A compliance team needs to process tens of thousands of PDF reports each night and generate concise GenAI summaries for auditors. The reports are uploaded to Amazon S3 throughout the day, and the summaries must be ready by the next morning. The team wants to avoid long synchronous API calls, handle failures for individual documents, and automatically retry only failed items without reprocessing the whole batch. How should the solutions architect orchestrate this large scale, batch GenAI summarization workflow?

A media platform uses generative models to produce large volumes of content recommendations and short summaries. Manual review for fairness and stereotyping has become impractical, but the trust and safety team still needs a consistent evaluation approach that identifies biased or unfair patterns across demographic groups with periodic human spot checks. As the responsible AI specialist, how would you design an automated judging system that evaluates fairness, reduces judge model bias, and remains aligned with human reviewers?

A SaaS company operates dozens of backend services that prepare prompts, validate requests, and post-process outputs for its GenAI platform. As the platform scaled, engineers introduced inconsistent coding patterns, weak test coverage, duplicated SDK logic, and inefficient API usage, leading to frequent regressions. Engineering leadership wants a standardized way to scaffold services, enforce internal architecture rules, automatically detect poor SDK usage, and generate reliable tests without increasing manual review overhead. What approach should the team adopt?

A multinational logistics company is launching a GenAI-powered customer inquiry assistant using Amazon Bedrock to handle shipment tracking, billing questions, and return requests. Compliance mandates that responses always follow a company-approved tone, avoid regulatory or legal guidance, and conform to a strict JSON schema consumed by downstream automation systems. Multiple microservices invoke the model across regions, and leadership wants centralized governance over prompts, behavior, and output format. How should the team design the instruction and governance layer to enforce these requirements consistently? (Select two)

A clinical decision support system uses a GenAI assistant to summarize treatment guidelines and suggest diagnostic pathways. Doctors complain that while the recommendations are useful, the assistant behaves like a black box and does not expose how conclusions are reached. They want insight into intermediate reasoning steps, confidence signals, and the logical flow behind each suggestion. How should the interaction pattern and UI be redesigned to improve transparency?

A university research group is experimenting with a state-of-the-art open-source language model that exceeds the memory capacity of any single GPU. The researchers require interactive inference access for experimentation and cannot reduce model size without losing critical reasoning capabilities. The model must be loaded across multiple devices and serve requests reliably. Which inference architecture should be used?

A biotech company trains generative models using sensitive genomic and clinical data across multiple AWS accounts for research, validation, and production. Regulators now require a complete lineage graph showing how each deployed model was produced, including datasets, preprocessing jobs, training runs, and model artifacts across all accounts. How should the company implement end-to-end lineage tracking?

A large enterprise runs multiple internal GenAI assistants for engineering documentation, HR policy lookup, and compliance analysis. Usage patterns vary significantly, with experimental teams frequently exceeding expected token volumes while production workloads must remain predictable and cost controlled. Finance leadership requires fine-grained chargeback by team and environment, enforceable token budgets, and controlled exceptions for high-priority business workflows. How should the GenAI platform team design token usage tracking and enforcement?

A global investment firm is building a real-time GenAI decision support system used by financial advisors, risk officers, and compliance reviewers. Each role requires different reasoning depth, safety constraints, and output formats. Client conversations happen live, so latency must remain low. The firm also needs centralized governance, predictable cost controls, and automated routing to the most appropriate foundation model per request. Which options best satisfy these requirements? (Select two)

A public sector policy office needs a GenAI drafting assistant that can synthesize statutes, regulations, and internal briefing notes into long-form policy drafts. All processing must remain inside AWS GovCloud, workloads are primarily asynchronous, and reviewers require grounded outputs with traceable sources. The team prefers managed services to minimize operational overhead and control costs. Which combination of services should be used? (Select TWO)

A biomedical research organization is enabling natural language analysis over longitudinal patient datasets for internal research. Regulations prohibit exposing direct identifiers or re-identifiable attributes to any foundation model. The solution must anonymize data before inference while preserving analytical usefulness for cohort-level studies. Which approaches should be combined? (Select two)

A regulated financial institution is deploying multiple foundation models for fraud detection, customer analytics, and credit risk scoring. Regulators require that each promoted model version automatically includes a complete, auditable model card documenting intended use, limitations, evaluation metrics, and risk ratings. Manual documentation is no longer acceptable. What should be implemented to ensure automated, regulator-ready model cards?

A procurement analytics company uses Amazon Bedrock to extract structured compliance indicators from supplier agreements. A new prompt version has been drafted to simplify output formatting, but the team wants to ensure that legally critical clauses detected by the previous prompt are not lost. Before approving the new prompt, they want an automated regression testing process that runs both prompt versions against the same historical dataset, stores structured results, and produces measurable comparison metrics over time. How should this regression testing workflow be designed using AWS services?

A research hospital exposes internal GenAI services to doctors through multiple existing applications that already authenticate users via a corporate identity provider. The security team wants to federate identities into AWS, enrich access tokens with role and department attributes, apply conditional access based on network location and time of access, and centralize authorization logic without changing current login flows. Which approach best meets these requirements?

A digital publishing platform is experimenting with different foundation models and prompt templates to generate headlines and article summaries. The product team wants to evaluate multiple variants simultaneously, capture quality and engagement metrics, automatically adjust traffic routing, and progressively roll out the highest performing configuration with minimal manual intervention. How should this automated experimentation and rollout system be implemented?

A data science team is fine tuning a custom text generation model using Amazon SageMaker. They want a fully managed pipeline that handles preprocessing, training, evaluation, conditional registration, deployment, and rollback if performance declines. The pipeline should be reusable and automatically triggered when new datasets or configuration updates are introduced. Which orchestration approach best supports this end-to-end workflow?

An online safety team runs scheduled moderation jobs that analyze large volumes of user-generated content with a foundation model. Each content item is currently processed individually, leading to poor GPU utilization and increased cost. The team wants to redesign the workload to improve throughput and ensure batch jobs consistently complete within their time window. How should the workload be redesigned to maximize utilization and throughput?

A global energy utility is modernizing its incident response process for power grid failures. Diagnosing outages requires correlating live telemetry feeds, reviewing historical repair records, and determining corrective actions across multiple teams. The company wants to use multiple AI agents with distinct expertise that can share insights, coordinate tasks, and dynamically assign work during investigations. As a GenAI developer, which approach should you implement to enable effective collaboration among specialized agents? (Select two)

A consulting firm offers a GenAI assistant to enterprise customers that answers questions using proprietary operating manuals, internal workflows, and confidential best-practice documents. Each customer requires strict isolation of their documents, guarantees that their data is never used to train foundation models, and responses that are grounded only in their own content. Which architecture best meets these privacy and grounding requirements?

An insurance provider is automating claim adjudication. Each claim requires evaluating policy rules, calling external coverage and billing systems, and reassessing eligibility after each data lookup until a final decision or escalation is reached. The system must support structured reasoning combined with repeated action execution. Which approach best enables iterative reasoning and action until a final decision is reached?

An enterprise compliance assistant ingests lengthy policy manuals and meeting transcripts. Users complain that responses frequently overlook the latest updates, which are often appended at the end of documents. The ingestion pipeline currently uses uniform chunk sizes without accounting for structure or recency. How should you redesign the chunking and retrieval strategy to improve answer accuracy?

A global publishing platform uses Amazon Bedrock to generate articles across multiple Regions. During traffic spikes, the platform occasionally encounters throttling errors for its selected model. The company wants to improve availability without changing application logic or switching to a different model family, while maintaining consistent outputs. Which configuration strategy best meets these requirements?

A pharmaceutical research company is preparing clinical trial datasets and physician notes to serve as a retrieval source for an Amazon Bedrock Knowledge Base. Certain patient identifiers and trial arm details must only be accessible to an internal compliance team, while anonymized statistics and summaries should be available to the broader research group. Which approach should the company implement?

An insurance company is building a GenAI-powered virtual assistant that handles claims, policy changes, and billing inquiries within a single chat interface. The company wants to automatically detect user intent and route each request to specialized downstream workflows that invoke different Amazon Bedrock models and prompts. The team prefers a machine learning–based approach over static rules. What is the best solution?

A GenAI-powered customer support platform handles thousands of conversations per hour. Engineers notice intermittent hallucinations and sudden latency spikes that are hard to reproduce. They suspect prompt drift, token usage anomalies, and downstream service variability. Which solution provides comprehensive end-to-end observability?

A consumer electronics company is building a “Knowledge Assistant” that helps product teams query user manuals, marketing brochures, teardown images, and call-center transcripts. The assistant must support multimodal retrieval, stream grounded responses with citations, route requests to different foundation models based on the team requesting information, and strictly isolate confidential pre-launch content. The company wants a fully managed solution without hosting models themselves. What should they do?

An online certification platform uses a multi-step GenAI workflow to generate practice questions, validate them, and publish them to students. Failures sometimes occur due to invalid inputs, throttling, or transient model errors. The platform needs automatic retries, dead-letter handling for unrecoverable failures, centralized logging, and visibility into execution paths. How should this workflow be orchestrated?

A large healthcare provider is introducing an AI system that helps generate clinical summaries by reviewing doctor notes, patient history, and recent interactions. These summaries must be produced reliably because they support real clinical decision making, but the compliance and operations teams are concerned about potential risks. They worry that the AI could enter repetitive reasoning cycles, continue processing far longer than intended, or fail unpredictably if supporting systems become unavailable during execution. They want a tightly controlled workflow that can automatically stop when necessary and prevent cascading failures during periods of instability. What do you recommend?

A media intelligence company processes millions of global news articles each day and provides real-time insights to analysts, advertisers, and policy research teams. To support fast and accurate information retrieval, the platform needs a vector search solution that can segment content by topic, combine semantic relevance with keyword precision, and accelerate lookup using approximate nearest neighbor search. The engineering team must also tune index mappings, shard counts, and replica strategies to handle query spikes during major events. Which solution should the team choose to meet these requirements?

A multinational bank is designing a compliance-focused document review assistant for its risk division. The tool must generate grounded explanations with paragraph-level citations, enforce blocked topics using safety controls, and ensure that any new prompt template undergoes a formal approval workflow before being deployed. The bank also mandates that all model invocations be retained for ten years due to regulatory laws and that the entire system operate with minimal custom infrastructure. Which of the following represents the best solution for the given use case?

A marketing content assistant recently adopted a new “tone and brand” prompt template that was manually tuned in a staging environment. After rollout, some product lines show improved copy while others regress, and the team has no structured way to compare the new template against the previous one across a broad test set. How should you introduce a prompt testing framework and evaluation workflow to systematically compare prompt versions and prevent quality regressions?

A B2B analytics vendor is modernizing its GenAI powered insights assistant on Amazon Bedrock. Business analysts currently submit free-form queries, and the foundation model responds with unstructured text that varies in quality and does not expose clear reasoning steps. The vendor wants to redesign the prompting approach by enforcing structured input fields, implementing strict JSON output schemas, applying chain-of-thought-style reasoning instructions, and capturing user feedback scores to iteratively refine prompt templates through a centralized workflow. As the GenAI architect, how would you enhance the prompt design and feedback loop to improve foundation model performance?

A digital publishing company operates a semantic search API that recommends “similar research papers” using a vector index stored in Amazon OpenSearch Service. As the corpus grows beyond 30 million embeddings, engineers notice query latency spikes during academic conference seasons. Monitoring reveals uneven shard utilization, and search relevance degrades for complex technical queries with dense semantic meaning. As the GenAI engineer, how should you troubleshoot and optimize the vector search setup to maintain consistent performance and relevance at scale? (Select two)

A document analysis workflow chains together multiple prompts that transform extracted text into structured JSON used by downstream compliance checks. Occasionally, later steps fail due to missing fields or unexpected formats, and engineers struggle to pinpoint which transformation introduced the issue. The team wants better traceability and early detection of schema violations across the pipeline. What should you recommend?

A multinational retail brand runs a conversational assistant that answers questions about products, warranties, and store policies. Usage analysis shows that many customers ask nearly identical questions with minor phrasing differences, yet each request triggers a full foundation model invocation. Customers in distant Regions also experience higher response times during flash sales. As the GenAI architect, how should you design a solution to reuse prior responses and reduce latency for global users? (Select two)

A content moderation platform analyzes images and captions using a single large multimodal foundation model. Most requests only require basic classification, but a small percentage need deep reasoning and explanation. Costs spike during peak traffic because every request invokes the expensive model. How should you redesign the workflow to optimize cost while preserving accuracy for complex cases?

A wealth management firm uses an Amazon Bedrock agent to generate personalized investment guidance. Auditors require full traceability into how recommendations are produced, including retrieval steps, tool calls, and intermediate reasoning, to support regulatory reviews. Which tracing and logging design best satisfies this requirement?