AI Infrastructure: Models, Connectors, and Runtime Foundations
What to Know
AI Infrastructure in Boomi covers the components that let processes call AI services and models. Candidates must understand connectors to AI providers (for example, LLM service connectors), model configuration (model selection, parameters such as temperature and token limits), prompt construction (templates, variable injection), and response handling (parsing model output into a profile). Understand the role of the Boomi runtime in hosting AI-enabled processes and the security considerations: credentials for AI providers must be stored as connection credentials, not hardcoded.
Know the integration patterns: synchronous request/response for real-time use cases, asynchronous for long-running inference, and batching for bulk processing. Understand error handling for AI calls - timeouts, rate limits, and malformed responses - and the need for retry and fallback logic. Understand data privacy: sensitive data sent to external models must be governed, and prompt injection is a real risk when user input is included in prompts.
Traps
A common trap is assuming AI connector calls are deterministic - they are not; responses vary. Questions may claim temperature controls output length (it controls randomness, not length; token limits control length). Another trap: assuming credentials can be embedded in the process - they must be in connection components. Watch for claims that AI calls are automatically retried; retry logic must be configured.
Rule of Thumb
If the requirement is real-time user-facing inference, use synchronous calls with timeouts and fallbacks. If the requirement is bulk classification, use asynchronous or batched processing with retry logic.
Worked Example
A process classifies support tickets using an LLM. Build a map that injects the ticket text into a prompt template, call the AI connector synchronously with a timeout, parse the response into a classification profile, and add a Try/Catch with a fallback classification if the call fails or times out.