In-Context Learning with Large Language Models for Scalable Glycemic Index Assignment to Food Composition Databases: Development, Validation, and Reproducibility

Assigning glycemic index (GI) values to food composition databases is a critical bottleneck in nutritional epidemiology. We developed an in-context learning approach using large language models (LLMs), in which a structured knowledge system (termed a skill) loads GI reference databases ([~]11,000 entries), expert decision rules, and error-correction heuristics into the models context window ([~]300,000 tokens). The LLM performs GI assignments without scripted logic, functioning simultaneously as a semantic matching engine, numerical reasoning system, and expert curator. We validated this approach in two experiments. In Validation Study 1, the skill predicted the expert-curated US National GI Database (9,428 foods) using only European reference data, achieving within {+/-}10 agreement of 73.7% without manual review - compared with 31.3% retention of previously published cosine-similarity approach. In Validation Study 2, the skill was augmented with US GIDB and applied to 1,157 European food descriptions classified using the EFSA FoodEx2 system, achieving ICC = 0.79 with the expert (weighted {kappa} = 0.65; triplicate ICC = 0.88). We then applied the skill prospectively to extend US dietary GI and GL surveillance to two additional NHANES cycles (2019-2023), identifying a continued decline in energy-adjusted glycemic load. Reproducibility was assessed through triplicate runs (temperature = 0, pinned model version). The skill architecture is described in sufficient detail to inform future applications of in-context learning for nutritional database construction. STATEMENT OF SIGNIFICANCEThis paper introduces a fundamentally new approach to glycemic index (GI) database construction. Rather than using programmatic text-matching algorithms followed by extensive manual curation, we demonstrate that a large language model (LLM), when loaded with the complete GI reference literature and formalized expert decision rules, can perform one-shot GI assignments at accuracy levels comparable to human expert ratings (ICC = 0.79 with expert, weighted {kappa} = 0.65 for GI category agreement). The approach is validated across two independent food databases spanning US and European food supplies. The method reduces the time required to assign GI values to a new national food database from months of expert labor to hours of computation, while maintaining reproducibility through a structured, versionable skill architecture. This has immediate practical implications for enabling GI-based dietary surveillance and epidemiologic research in countries that currently lack GI databases or need to update existing databases.