NutriMed AI

• 

  NutriMed AI Multi-Agent System in action-safely delivering personalised dietary recomendations.

Inspiration

In today’s fast-paced world, managing chronic conditions like diabetes and hypertension while juggling medications is incredibly challenging. I watched my grandmother constantly worry about which foods would interact with her Lisinopril or Metformin. Generic diet apps failed her — they ignored drug interactions, allergies, and real clinical context. This frustration sparked the idea: What if we built a team of AI agents that functioned like a complete clinical nutrition team — always available, deeply personalized, and safety-first? Thus, Agents Assemble: The Healthcare AI Endgame (NutriMed AI) was born.

What it does

NutriMed AI is a multi-agent healthcare system that delivers safe, intelligent, and personalized dietary guidance. It acts as a collaborative team of specialized AI agents that: Pulls real patient data (via FHIR) Checks allergies and medical history Provides accurate dietary recommendations Detects dangerous food-drug interactions Evaluates herbal supplements in full clinical context Generates realistic meal plans The agents communicate with each other (A2A), debate risks, and only deliver final recommendations after cross-verification — mimicking how a real doctor + dietician + pharmacist would work together.

How we built it

We created seven specialized tools and integrated them into a coordinated multi-agent system: GetPatientId + GetPatientAllergies GetDietaryRecommendations CheckFoodDrugInteraction GetHerbalSupplement GetMealPlan These tools are orchestrated using agent-to-agent communication, allowing dynamic consultation. The system pulls live patient context, runs safety checks, and delivers warm, actionable advice. We used structured prompting, tool calling, and a supervisor layer to maintain reliability and clinical safety.

Challenges we ran into

Coordinating multiple agents without contradictory advice was difficult. Early versions had agents going in different directions. Ensuring clinical safety while keeping responses helpful and natural. Handling complex multi-agent workflows (A2A calls) and debugging integration issues between tools. Balancing creativity in meal planning with strict medical constraints.

Accomplishments that we're proud of

Successfully built and integrated all 7 tools working together seamlessly. Achieved real clinical intelligence: the agent correctly identified a patient with Type 2 Diabetes + Hypertension on Metformin + Lisinopril, gave accurate interaction warnings, and provided responsible advice. Implemented working Agent-to-Agent (A2A) communication and FHIR context handling. Delivered a system that feels both intelligent and empathetic.

What we learned

We learned that multi-agent systems are incredibly powerful but require strong orchestration and safety guardrails. Most importantly, we realized the future of healthcare AI lies in collaboration — not replacing doctors, but augmenting them with scalable, intelligent support systems that can reach millions of patients.

What's next for NutriMed AI

We plan to expand NutriMed AI with voice interaction, wearable data integration, cultural recipe adaptation across more cuisines, and deeper integration with hospital systems. Our ultimate goal is to make world-class personalized nutrition support accessible to everyone.

Built With

• fastapi
• gemini3.1flashlite
• mcp
• promptopinion
• python
• uvicorn