Little Lunches is a fast-growing personalized meal planning app trusted by over 500,000 parents globally. The platform provides curated, age-appropriate recipes and feeding guidance developed by certified dietitians, pediatricians, and feeding therapists.
Problem: scaling personalized nutrition with an AI nutrition assistant
As Little Lunches scaled, parents increasingly sought instant answers to child and family nutrition questions. Without an AI nutrition assistant, the platform struggled to deliver expert guidance at scale, relied on static content with limited personalization, and faced growing manual effort to support multilingual users.
At the same time, IP risks from exposing proprietary content and the lack of a structured data foundation limited long-term personalization. To overcome this, Little Lunches needed a secure, multilingual AI nutrition assistant built for healthcare and family nutrition.
Solution: an Agentic AI-powered nutrition assistant
GoML designed and delivered an MVP AI-powered dietitian assistant using GoML’s Agentic AI, purpose-built for healthcare and nutrition use cases.
This AI nutrition assistant enables parents to ask natural language questions and receive empathetic, context-aware nutrition guidance while dynamically suggesting relevant recipes, articles, and feeding guides using metadata-only retrieval.
The system combines agentic LLM-based reasoning, vector-powered content retrieval, and anonymized data logging to ensure scalability, compliance, and long-term personalization of the AI nutrition assistant.
Conversational experience powered by an AI nutrition assistant
Parents interact with a warm, non-judgmental AI nutrition assistant directly within the Little Lunches app.
- Supports English and Spanish conversations
- Reflects the persona of certified dietitians and feeding therapists
- Uses child age, allergies, and dietary preferences for personalization
- Suggests actionable follow-ups such as “Make this dairy-free” or “Add to meal plan”
The AI nutrition assistant ensures guidance is accurate, empathetic, and age-appropriate.
Secure LLM orchestration for the AI nutrition assistant
User prompts are securely routed to Claude Sonnet 3.5 via Amazon Bedrock, enabling the AI nutrition assistant to reason contextually while remaining domain-restricted.
- User context is injected securely at runtime
- Strict guardrails limit responses to nutrition and feeding topics
- Profanity and unsupported topics are filtered
- The AI nutrition assistant never accesses full proprietary content
Vector-based content retrieval for the AI nutrition assistant
To power intelligent recommendations, the AI nutrition assistant uses a pgvector + PostgreSQL semantic search layer.
- Recipes, articles, and feeding guides are embedded as vectors
- Only metadata (titles, summaries, tags, IDs) is retrieved
- Relevant content is displayed inline below AI responses
- Proprietary content remains fully protected
Anonymized data logging to improve the AI nutrition assistant
Every interaction with the AI nutrition assistant is logged securely.
- Prompts, responses, language, dietary tags, and topics are captured
- PII is anonymized before storage
- Data is exported to an AWS S3 data lake
- Logs create a foundation for future fine-tuning of the AI nutrition assistant
Impact of the AI nutrition assistant
- Scaled expert nutrition guidance through an AI nutrition assistant
- ≥90% successful recipe/article match rate
- Sub-4 second response latency for AI nutrition assistant interactions
About
Before Gen AI and after Gen AI
“For Little Lunches’ AI-powered dietitian assistant, GoML transformed its static nutrition content into a secure, empathetic, and scalable conversational experience without compromising data privacy or proprietary assets.”
Prashanna Rao, Head of Engineering, GoML
Key takeaways for digital health and nutrition platforms
Common challenges
- Personalized guidance does not scale with manual workflows
- Multilingual support increases operational complexity
- Uncontrolled LLM access risks IP and compliance
Practical guidance
- Use a vector-based metadata layer to protect proprietary content
- Inject user context securely at runtime
- Design anonymized data pipelines from day one
Ready to build a secure AI-powered assistant?
Partner with GoML to accelerate the development of production-ready AI systems with AI Matic.
.png)
