What you should be able to do
- Translate customer language into structured preferences.
- Design explainable recommendation logic.
- Recognize AI hallucination and overconfidence risk.
- Apply privacy-minimizing personalization.
AI should reduce menu friction
A useful AI Barista asks a small number of meaningful questions—strength, milk, sweetness, temperature, flavor direction, caffeine tolerance or mood—and converts them into ranked options. The customer should understand why each option was suggested.
Structured preference model
Store only what is useful: for example intensity, sweetness, dairy preference, hot/iced, flavor family and prior likes. Avoid collecting sensitive or unnecessary personal information simply because the interface can.
Explainability
Recommendations should point to observable menu attributes: “You asked for strong, low-sweetness and iced, so this drink ranks higher.” Avoid opaque “AI says this is perfect for you” language.
Hallucination controls
The model must not invent ingredients, allergens, prices, availability or health claims. Ground recommendations in an approved menu dataset and validate critical fields in application logic rather than trusting free-form generation.
Human override
Baristas should be able to correct recommendations based on stock, allergies, service context or customer clarification. AI assists the conversation; it does not own the customer relationship.
Privacy by design
Make preference storage optional where possible, define retention, protect profiles, and provide a simple way to clear saved preferences. Personalized coffee does not require surveillance.
Apply the lesson
Write a six-question AI Barista intake that captures enough information to rank drinks without asking for name, phone number or other unnecessary identity data. Then write one explainable recommendation response.
Open Coffee Lab ↗Test your understanding
Pass at 70% or higher, then mark the module complete. You can retry as many times as needed.
Progress is stored locally in this browser. Completing a module does not issue a formal certification.