(PTFS06-03-26) NutriOrion: A Safety-Constrained Multi-Agent AI Framework for Scalable Personalized Nutrition Consultation

Sunday, July 26, 2026

11:08 AM - 11:12 AM ET

Location: Prince George's Hall B-E (The Hub) - Poster Theater 2 / Poster Board Number: PT2-3

Emerging Leaders in Nutrition Science Poster Competition

Presenting Author(s)

Junwei Wu, B.Eng

Graduate student
Emory University
Decatur, Georgia, United States

Disclosure(s): No relevant financial relationship(s) with ineligible companies to disclose.

Co-Author(s)

Hanqi Luo, PhD

Assistant Research Professor
Emory University
Oakland, Georgia, United States
Darren Liu, MS

Graduate Student
Emory University
Atlanta, Georgia, United States
MW

Minxiao Wang

Emory University, United States
KT

Kimberly L. Townsend, MS

Graduate Student
Texas Woman's University
Conroe, Texas, United States
LH

Lydia Hartwig, MS, RD

Cecelia Health
Beale AFB, California, United States
Derek C. Miketinas, PhD, RD

Associate Professor
Texas Woman's University
Houston, Texas, United States
XH

Xiao Hu, PhD

Emory University, United States
CY

Carl Yang

Emory University, United States
Runze Yan, PhD

Assistant Professor
Emory University
Atlanta, Georgia, United States

Disclosure(s):

Junwei Wu, B.Eng: No relevant financial relationship(s) with ineligible companies to disclose.

Darren Liu, MS: No relevant financial relationship(s) with ineligible companies to disclose.

Derek C. Miketinas, PhD, RD: Grain Food Foundation: Grant/Research Support (Ongoing)
Ingredion, Inc.: Grant/Research Support (Ongoing)
Mead Johnson Nutrition/Reckitt: Grant/Research Support (Ongoing)
National Cattleman's Beef Association: Grant/Research Support (Ongoing)

Runze Yan, PhD: No relevant financial relationship(s) with ineligible companies to disclose.

Objectives: Personalized nutrition care for stroke patients is challenged by multiple comorbidities, medication-related constraints, and complex dietary guidelines. We developed NutriOrion, a hierarchical multi-agent large language model (LLM) framework designed to provide clinically grounded, drug–nutrient interaction (DNI)-safe dietary recommendations.

Methods: NutriOrion integrates three specialized analytical modules (disease-specific, medication-aware, and nutrient optimization). We evaluated NutriOrion using data from stroke patients (n = 330) from NHANES 2019–2021, constructing each participant's profile from: (1) Day-1 24-hour dietary recall; (2) medical diagnoses; and (3) prescription medications. NutriOrion’s recommendations were evaluated across four domains: 1) recommendation quality via NCP-QUEST (0–8; AHA/ADA guideline adherence, clinical appropriateness, personalization, feasibility), assessed by two dietitians blinded to source (Cohen's κ); 2) actionability and 3) DNI safety — both compared against general-purpose GPT and Claude; and 4) nutritional impact via 1:1 meal substitution, with macro- and micronutrients recalculated.

Results: NutriOrion performed well across all four evaluation domains. 1) Clinician-facing quality: NutriOrion's recommendations achieved a mean NCP-QUEST score of 7.5/8 (inter-rater agreement κ = 0.82). 2) Actionability: NutriOrion's recommendations achieved 97.8% actionability, compared with 91.2% for both GPT and Claude. 3) DNI safety: NutriOrion's recommendations had a DNI violation rate of 12.1%, compared with 16.6% (GPT) and 11.5% (Claude). 4) Nutritional impact: substitution analysis demonstrated favorable dietary improvements — +167% fiber, +27% potassium, −9% sodium, and −12% added sugars following NutriOrion — compared with baseline intake.

Conclusions: NutriOrion generated personalized, medication-aware dietary recommendations that dietitians rated as actionable and guideline-adherent, with fewer harmful food-drug combinations than general-purpose AI tools. These findings support its potential to scale up to provide individualized nutrition care.

Funding Sources: No external funding