Exam Scenarios

A fintech startup wants to add an AI chatbot for customer support. They have sensitive financial data, need low latency, and want to start with a PoC before going to production. How do you design this?

A multinational company needs their RAG system to handle queries in English, German, and Japanese. Documents are in all three languages. The system must return results in the user's language. Design the retrieval pipeline.

Your production GenAI app serves users in US and Europe. The primary model (Claude) occasionally hits rate limits during peak hours. How do you ensure availability?

A healthcare company has domain-specific medical terminology that general FMs handle poorly. They need to fine-tune a model on their proprietary medical corpus. Which path?

An insurance company receives claims as scanned PDFs containing handwritten notes, printed text, and attached photos of vehicle damage. They want to extract all information and generate a structured summary. Design the pipeline.

An e-commerce company wants semantic search across 2 million product descriptions AND exact-match filtering by product SKU, price range, and category. Users also need analytics on what they're searching for. Which vector store?

A legal firm's RAG system retrieves contract clauses, but users complain that retrieved chunks often split mid-clause, producing incoherent context. The chunking uses 512-token fixed-size. How do you fix this?

A company has 15 different GenAI applications, each with their own prompts stored in application code. Prompt changes require code deployments. The compliance team wants audit trails for all prompt changes. How do you centralize this?

A DevOps team wants to build an agent that monitors their Kubernetes cluster, reads alerts from SigNoz, queries Prometheus, and suggests remediation actions. The agent should use their existing Python monitoring scripts. Which architecture?

A financial analyst needs an agent that can download CSV files from internal APIs, perform statistical analysis, and generate visualizations. The code execution must be sandboxed. Which architecture?

An enterprise has 10,000 daily FM API calls. 70% are simple FAQ lookups, 20% are moderate analysis, 10% are complex reasoning tasks. Current cost is $X/month using Claude Sonnet for everything. How do you optimize?

A company wants to build an MCP server that exposes their CRM data to multiple AI agents built with different frameworks (Strands, LangGraph). Some tools need persistent database connections. How do you implement this?

A development team wants to automatically test prompt quality before deploying prompt changes to production. They use CodePipeline for CI/CD. Design the pipeline.

A customer service chatbot accidentally reveals another customer's email address in its response. The compliance team demands immediate remediation. What do you implement?

An insurance company needs to verify that policy quotes generated by their FM are mathematically correct and comply with their rate tables. Contextual grounding alone is insufficient because the FM needs to do calculations. What do you implement?

A regulated financial institution wants to use Bedrock but requires: (a) no data traversing public internet, (b) audit trail for every model invocation, (c) encryption at rest with customer-managed keys, (d) only specific teams can use specific models. Design the security architecture.

An HR department deploys a resume screening agent. The legal team raises concerns about demographic bias. How do you address this?

A startup's GenAI app costs $12,000/month on Bedrock. Analysis shows 60% of queries are near-duplicates (users asking the same thing in different words). How do you cut costs?

After a prompt update, users report that the RAG chatbot is 'making things up' -- responses contain information not in the knowledge base. How do you diagnose and fix this?