An AI waiter for Boxy Park — diners describe what they feel like eating in plain language ("suggest something for date night, I'm vegetarian"), Allen asks one or two clarifying questions about spice and allergies, then recommends 2–4 real menu dishes filtered by their dietary needs and constraints.

6-MenuMate.png

Live App: https://ask-allen.lovable.app/ Built by: Snigdha Babbar


Product Thinking

👤 Who is your user?

Restaurant diners at Boxy Park who want personalized dish recommendations but can't afford a knowledgeable waiter's full attention — especially solo diners, dietary-restricted guests, and indecisive groups.

🔴 What problem does your product solve?

Menus are static lists. Diners with constraints (allergies, dietary preferences, budget, mood) waste 5-10 min scanning, asking servers, or defaulting to safe-but-boring choices. Servers are too busy to give every table a thoughtful 5-minute consultation.

⚡ Core flow (User does X → System does Y → User gets Z)

User Action System Step 1 System Step 2 User Gets
"Suggest something for date night, I'm vegetarian" Allen detects dietary intent ("vegetarian") + retrieves only veg/vegan dishes from menu via embedding similarity Allen asks 1-2 clarifying questions (spice, allergies), then recommends 2-4 dishes grounded ONLY in the retrieved menu chunks A short list of real menu dishes that match their stated constraints, with one-line reasons + a drink-pairing offer

AI Architecture

System Prompt Design — Components

Persona "Allen, a calm and attentive waiter at Boxy Park" — warm, patient, never robotic, never pushy. Chosen because waiters embody trust + restraint, opposite of pushy chatbots.

Guardrails Rules 1-7: never invent dishes (Rule 1); always check dietary/allergens (Rule 2); never list excluded dishes even with disclaimers (Rule 3); vegan ≠ vegetarian, exclude honey/dairy/eggs by description (Rule 4); never guess — say "I don't know" honestly (Rule 5); never push expensive (Rule 6); answer-to-question alignment for ambiguous replies (Rule 7).

Output Format 1-2 line warm intro → bulleted picks (Dish — ₹Price — one-line reason) → drink pairing offer if food was recommended → 2-4 dishes per response.

Tone Calm, warm, professional. Friendly without being chatty. Food advice first, focused replies.