AI Nutritionist & Automated Meal Planning

Client

gbMeals

Year

2025

Location

UK

Domain

Generative AI

To solve this challenge, we designed a multi-stage AI pipeline that combines structured data processing, controlled LLM generation, and automated document production.

The process begins with a structured preference ingestion layer that captures user goals, body metrics, dietary restrictions, cooking habits, and ingredient preferences through a detailed intake form. Rather than sending raw responses directly to the model, inputs are normalized into a structured schema defining caloric targets, macro ranges, excluded ingredients, and meal-prep preferences. This structure ensures downstream systems can enforce strict constraints while keeping the generation process flexible.

At the core of the system is a specialized meal-planning LLM that generates complete weekly meal plans. The model produces both everyday cooking recipes and optional bulk meal-prep strategies for users who prefer cooking once for the week. Prompt engineering and domain-specific instructions guide the model to create realistic recipes, ingredient lists, and preparation steps suited for everyday home kitchens.

To ensure reliability, we implemented anti-hallucination guardrails around the generation process. These validation layers check outputs against dietary restrictions, ingredient safety rules, and realistic quantity ranges. If the model generates an inconsistency, the system automatically corrects or regenerates the output, ensuring the final plan behaves more like a dependable virtual nutritionist than an unconstrained text generator.

Once validated, the plan moves through an automated document generation pipeline that formats recipes, preparation instructions, and consolidated shopping lists into a clean, modern PDF. The personalized meal plan is then automatically delivered to users via email on a recurring schedule, allowing gbMeals to scale personalized nutrition while maintaining quality and dietary safety.