Isadora Martin-Dye describes a four-layer design for customer-facing AI: identity constraints, situational context, example-based voicePrompt engineering is the practice of designing and refining instructions, context, examples, and constraints to obtain useful AI outputs. and a post-generation checkEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. She explains why a single tone instruction may fail on situations outside its examples.
Using wedding venue and missing-persons product examples, she discusses AI disclosure, avoiding claims of physical presence and adapting tone to sensitive situations. She presents these as rules chosen for her own products.
She describes checking generated dates and numbers against allowed values, and requiring explicit tenant identity rather than a silent default. She acknowledges that deterministic pattern checks cover only the cases they are designed to catch.
Watch on YouTube
