Integrating AI into Enterprise Systems CAIPM Practice Questions (Page 9)

Questions 41–45

1. 41foundation · easy

   Which approach is commonly used to enable AI systems to access data from legacy databases that lack modern APIs?

   • AManually exporting data from the legacy system to CSV files and importing them into the AI system
   • BUsing database connectors or ODBC/JDBC drivers to interface directly with the legacy database
   • CUsing screen scraping to capture data from the legacy application's user interface
   • DReplacing the legacy database with a modern cloud database
   Reveal Solution Discussion Select an answer first
2. 42expert · hard

   A logistics company is integrating an AI model for delivery route optimization. The model uses real-time traffic data, weather data, and historical delivery data. The historical data is stored in a legacy data warehouse that is updated nightly. The real-time data comes from third-party APIs. The company wants to ensure the model's predictions are accurate and that the integration can handle API failures. What should the data integration strategy include?

   • AUse only the historical warehouse data and update the model daily, ignoring real-time traffic and weather.
   • BUse a data pipeline that combines nightly batch data from the warehouse with real-time API data, and implement fallback logic to use the last known good data if an API fails.
   • CUse only the real-time API data and ignore the historical warehouse data to simplify the integration.
   • DIngest all data in real time by forcing the legacy warehouse to provide real-time updates.
   Reveal Solution Discussion Select an answer first
4. 43foundation · easy

   When designing the data flow for an AI component in an enterprise system, which sequence best represents the typical flow?

   • AData preprocessing → model training → data ingestion → output to application
   • BOutput to application → model inference → data preprocessing → data ingestion
   • CModel training → data ingestion → output to application → data preprocessing
   • DData ingestion → data preprocessing → model inference → output to application
   Reveal Solution Discussion Select an answer first
5. 44expert · hard

   A logistics company is integrating an AI route-optimization service into its dispatch system. The service is deployed as a microservice and must handle peak loads during holiday seasons. The team has limited budget and cannot over-provision infrastructure. What should the integration architect do to ensure the system can handle peak loads without excessive cost?

   • AUse an autoscaling group that scales the microservice based on CPU utilization or request queue depth.
   • BUse a queue to buffer requests and process them slowly during peak periods.
   • CDeploy the microservice on a fixed number of virtual machines sized for peak load.
   • DMove the microservice to a serverless function with a maximum execution time of 15 minutes.
   Reveal Solution Discussion Select an answer first
6. 45expert · hard

   A global retail company is integrating an AI-based demand forecasting system. The company operates in regions with varying data residency laws and has a mix of on-premises data centers and cloud subscriptions. The forecasting model must be updated daily with sales data from all regions. The company wants to minimize data transfer across borders while maintaining a single, consistent model. What should the AI program manager recommend?

   • AUse a single cloud region that has the most permissive data laws and train the model there.
   • BDeploy the model in each region using local data for training, and then aggregate the models into a global ensemble.
   • CUse a federated learning approach where models are trained locally in each region and only model updates (not raw data) are shared to build a global model.
   • DCentralize all data in the company's headquarters and train the model there, ignoring regional data residency laws.
   Reveal Solution Discussion Select an answer first