Nutrimex

The Problem
Mexico has one of the highest rates of diabetes and obesity in the world — yet access to a certified nutritionist remains expensive, scarce outside major cities, and impractical for everyday decisions. NutriMex closes that gap with an AI assistant that combines real nutritional knowledge with prices from the convenience stores Mexicans actually shop at.

What NutriMex Does
NutriMex is an agentic generative UI interface that acts as a pocket nutritionist. Users chat in natural language and the agent:
Generates personalized weekly meal plans by health goal (diabetes control, weight loss, better nutrition, weight maintenance) and city
Compares food basket prices across OXXO, 7-Eleven, Walmart Express, and Soriana — recommending the most affordable option
Checks COFEPRIS status of products (Mexico’s NOM-051 nutritional labeling standard) to flag items with excess sugar, sodium, or fat
Builds an optimized shopping cart with health warnings included
Queries a real nutritionist knowledge base hosted in Notion (markdown pages authored by nutritionists, with embedded food databases) to answer clinical questions from actual sources — not hallucinated data

Generative UI in Action
NutriMex uses the full Generative UI spectrum:
Level
Usage
Controlled (useComponent)
Interactive weekly plan cards, shopping ticket, and macronutrient charts — the agent picks and populates the right component
Declarative (A2UI)
Price comparison components streamed dynamically by Gemini with no executable code on the client
Open-ended (MCP Apps)
The NutriMex MCP server runs natively inside Claude and ChatGPT — same agent, three surfaces

Tech Stack
Layer
Technology
Frontend
Next.js 15 + React 19 + TailwindCSS 4 + Shadcn/ui
AI Agent
LangGraph Deep Agents + Gemini 3.1 Flash-Lite (swappable with Claude Sonnet 4.6)
Orchestration
CopilotKit Intelligence (durable threads, Postgres + Redis)
Knowledge Base
Notion (markdown pages + embedded databases, read via @notionhq/notion-mcp-server)
MCP Server
mcp-use (TypeScript), deployable to Manufact Cloud
Regulatory Data
COFEPRIS / NOM-051 (local nutritional labeling database)
BFF
Hono (TypeScript)

What Makes It Different
Real knowledge, not hallucinations. Before answering, the agent searches and reads Notion pages written by nutritionists — using three tools (search_nutrition_kb, read_nutrition_page, read_nutrition_db) that convert markdown pages and embedded tables into structured LLM context.
Mexican context from the ground up. Meal plans use local ingredients (nopal, black beans, chia, jicama). Price comparisons cover the country’s most accessible convenience store chains. Health alerts cite COFEPRIS/NOM-051 regulations.
Price as a health variable. Health isn’t just calories — it’s whether the user can afford the basket. NutriMex integrates price, accessibility, and nutritional quality into a single recommendation.
Deployable inside Claude and ChatGPT. The NutriMex MCP server can be added as an MCP App in any compatible client, bringing the tools to where users already are.

Potential Impact
With ~14 million people living with diabetes and ~75% of adults overweight or obese in Mexico, NutriMex targets one of the country’s most pressing public health challenges — with a tool that is accessible, free, and works from a phone.