Optimal Meal Schedule for a Local Nonprofit Using LLM-Aided Data Extraction

TL;DR

The paper tackles local food insecurity by enabling cost-conscious, nutritionally adequate meal planning for a nonprofit. It combines PDF data extraction, LLM-based ingredient standardization, and an inflation-aware cost model with a binary linear program that minimizes $\sum_{j=1}^N R_j x_j$ subject to $x_j \in \{0,1\}$ and nutrition/category constraints. Key contributions include mapping 866 ingredient items across 157 recipes to the My Food Data database, deriving per-recipe costs, and generating a 15-week optimized meal plan that balances cost and nutrition while supporting real-time decision-making via a web platform. The result is a practical, extensible workflow that can be adapted by other nonprofits to improve ongoing meal planning and budgeting under price volatility.

Abstract

We present a data-driven pipeline developed in collaboration with the Power Packs Project, a nonprofit addressing food insecurity in local communities. The system integrates data extraction from PDFs, large language models for ingredient standardization, and binary integer programming to generate a 15-week recipe schedule that minimizes projected wholesale costs while meeting nutritional constraints. All 157 recipes were mapped to a nutritional database and assigned estimated and predicted costs using historical invoice data and category-specific inflation adjustments. The model effectively handles real-world price volatility and is structured for easy updates as new recipes or cost data become available. Optimization results show that constraint-based selection yields nutritionally balanced and cost-efficient plans under uncertainty. To facilitate real-time decision-making, we deployed a searchable web platform that integrates analytical models into daily operations by enabling staff to explore recipes by ingredient, category, or through an optimized meal plan.

Figures (4)

• Figure 1: Recipe optimization workflow. A: Excerpt from one recipe used for data extraction. B: Flowchart of our approach, starting from the provided recipes and invoices, to our optimal recipe recommendations.
• Figure 2: Matching strings of PPP recipe ingredients to the My Food Database. A: projection of five phrases onto the first two principal components of the SentenceTransformer vector space. B-C: Successful ingredient matches after stage one (B) or stage two (C). Orange highlighting indicates manually identified best match.
• Figure 3: Nutritional Metrics of PPP Recipes. A-B: Projections of PPP recipes onto Fat vs. Calories, and Protein vs. Calcium. Dashed lines indicate the bounds for average fat, calcium, and protein during optimization; filled circles indicate recipes included in the optimal schedule, C-E: Distribution of other dietary elements, separated by protein category.
• Figure 4: Price Estimation and Prediction.(A) Historical prices per pound of selected ingredients. (B) Predicted recipe costs by protein category. (C) Histogram of percentage increase in predicted cost across recipes.