COOKING WITH PURPOSE: A COMPREHENSIVE FEASIBILITY AND SCALABILITY ASSESSMENT USING LARGE-LANGUAGE-MODEL-ASSISTED EVALUATION

Abstract

Background: Youth in South-side Tucson, a designated food desert, experience elevated rates of hypertension, type 2 diabetes, and obesity. ‘Cooking with Purpose' is a culinary education program designed to equip participants with practical cooking skills and nutritional literacy. Objective: To evaluate the feasibility and potential scalability of the pilot program using a large-language-model (LLM) pipeline that converts qualitative feedback into quantitative metrics. Methods: Twenty participants (aged 6-20) completed three 90-minute sessions delivered bi-weekly at a public library. Pre- and post- questionnaires (2-3 items each, Likert + open-ended) captured engagement, satisfaction, and behavior-change intent. GPT-4 (temperature 0.0) scored free-text responses on a 1-5 rubric. Nested averaging aggregated item-level scores to participant-level and cohort-level means. Feasibility criteria were ≥80 % completion, post-program satisfaction ≥4.0, and behavior-change intent ≥4.0. Results: All 20 enrolled youths completed every session (100% retention). Post-program satisfaction averaged 4.6±0.5, and GPT-4-derived behavior-change intent averaged 4.2±0.6, surpassing feasibility thresholds. Automated scoring processed all surveys in 5 min (<15 s per participant). Conclusions: High retention and strong engagement indicate that ‘Cooking with Purpose' is feasible for underserved youth. The LLM-powered evaluation pipeline markedly reduces analytic burden, supporting cost-efficient scale-up. Longer-term follow-up and multi-site trials are warranted.