Fourteen million Americans have food allergies. Two hundred thousand require emergency hospital treatment every year due to accidental allergen exposure in restaurants. Behind every one of those incidents is the same failure chain: a guest who communicated their allergy, a server who didn't record it accurately, and a kitchen that wasn't properly informed.
AI-powered allergen management eliminates two of those three failure points. According to Springer's International Journal of Information Technology (2024), multi-agent AI systems achieve 98.5% accuracy in allergen identification at the ingredient level — a dramatic improvement over manual staff tracking, which research estimates at 60-70% in high-pressure service environments.
The FASTER Act (2023) expanded the list of major food allergens in the United States from 8 to 9. These must be tracked at the ingredient level — not just flagged at the dish level — because cross-contamination can introduce allergens even into dishes that don't contain them as direct ingredients.
FlashBook's Doctor Agent implements a sequential 4-filter pipeline that processes every reservation where dietary information has been declared. Each filter builds on the previous, ensuring no allergen risk reaches the kitchen or the guest undetected.
Removes all menu items containing declared allergens from the recommendation pool. This filter is absolute — no exceptions, no overrides. If a guest declares a shellfish allergy, every dish containing any shellfish ingredient is removed regardless of preparation method.
Flags dishes prepared on shared surfaces, in shared fryer oil, or with shared utensils that also contact allergen-containing ingredients. A french fry technically contains no shellfish, but if it's fried in the same oil as battered shrimp, it carries a cross-contamination risk that this filter catches.
Applies structured dietary constraints beyond allergens: vegan (no animal products), vegetarian (no meat), halal (no pork/alcohol), kosher (no pork/shellfish, no mixing meat and dairy), and gluten-free. Each constraint is applied as a boolean filter against the ingredient database.
Ranks the remaining safe items by BCG quadrant (Stars first), guest preference history, and margin optimization. The output is a personalized safe menu — not a generic allergen-free list, but a curated recommendation that balances safety, preference, and profitability.
Managing allergens for a single guest is straightforward. Managing them for a group of 8 where three guests have different restrictions — one shellfish allergy, one vegan, one gluten-free — requires computing the intersection of all constraints to find dishes safe for everyone at the table.
"For group dining, the AI computes a shared-plate recommendation set that satisfies all constraints simultaneously. A dish is only recommended for the table if it passes every individual guest's allergen and dietary filters." — FlashBook Doctor Agent architecture
This is computationally trivial for an AI system but cognitively demanding for a server managing a busy service. The group dining intersection problem is one of the clearest examples of AI providing genuine safety value — not convenience, but risk reduction.
One principle is non-negotiable regardless of AI accuracy: the Doctor Agent always includes a disclaimer directing severe allergy cases to speak with kitchen staff directly. A 98.5% accuracy rate is exceptional. A 1.5% error rate on a severe peanut allergy is a hospitalisation. AI facilitates the safety conversation — it never replaces it.
Every recommendation output from the Doctor Agent includes: "For severe allergies or anaphylaxis risk, please speak directly with our kitchen team before ordering."
The Doctor Agent tracks Big 9 allergens per ingredient, detects cross-contamination risk, and generates personalized safe menus for every booking.
See FlashBook IA