Challenges
Need for personalized meal recommendations with clinical compliance
Healthcare dining providers must balance patient preferences with prescribed diets, allergy profiles, nutrient limits, standing orders, menu availability, and facility-specific rules. Selecting the most appropriate meal can be complex and time-consuming for patients and caregivers.
The client needed one personalized recommendation per patient and meal, full clinical and nutritional compliance, lower cognitive effort, natural-language refinement, and greater dietary variety without compromising clinical safety.
Solution
Personalizing Patient Meal Choices While Maintaining Clinical Compliance
Zensar built a cloud-native, AI-powered recommendation platform on AWS that combines generative AI with deterministic validation.
Amazon Bedrock generates and ranks meal candidates using patient, dietary, menu, and behavioral data. Every candidate passes an independent validation layer that checks nutrient, purchase, standing-order, availability, and food-group rules before an approved recommendation is presented to patients or caregivers. Prompt templates are version-controlled for consistent governance and review.
Background data preparation retrieves patient demographics, historical ordering patterns, and peer cohort information from the data warehouse. Session-level and per-day caching minimizes repeated warehouse queries and reduces latency.
Amazon Bedrock generates multiple meal candidates using dietary prescriptions, allergy profiles, standing orders, nutritional limits, historical ordering behavior, daily menu availability, and peer cohort preferences.
AI-generated candidates are evaluated and ranked, but never accepted directly. A deterministic validation layer checks nutrient, purchase, standing-order, availability, and food-group rules before a recommendation is shown.
Natural-language refinement lets users iteratively improve recommendations. Prompt templates are version-controlled and maintained using reusable YAML configurations to support governance and review.
Solution enablers
Tech Stack:
Generative AI
Deterministic validation
Cloud-native architecture
Containerized application
Version-controlled YAML prompt configurations
Automated CI/CD
Zensar Services:
AI solution design
Cloud-native implementation
AWS Services:
Amazon Bedrock
AWS Identity and Access Management (IAM)
Amazon ECS on AWS Fargate
Impact
Simpler meal ordering with measurable performance at scale
95% recommendation success rate during load testing
8.9-second median response latency across load testing
1.79 requests per second throughput demonstrated in testing
Reduced warehouse queries through intelligent caching
Business outcome
By simplifying patient meal selection, the solution reduced caregiver workload and enabled personalized dining services without compromising clinical compliance.
Conclusion
The implementation shows how generative AI can support clinical decision support with deterministic guardrails. Amazon Bedrock, IAM security, and governed prompts help create a trusted foundation.