PharmAssist AI

💡 Inspiration

Retail pharmacies are under more pressure than ever. Pharmacists spend hours managing supply chains, cross-checking stock, and handling repetitive administrative tasks, leaving them with very little time for actual patient care. On the other side, consumers navigate crowded aisles looking for OTC wellness products with little to no guidance.

We built PharmAssist AI to bridge this gap. By deploying multi-agent AI systems, we wanted to streamline the heavy backend logistics for retail pharmacies while simultaneously offering a highly personalized, intelligent wellness concierge for the everyday consumer.

⚙️ What it does

PharmAssist AI is an agentic, dual-purpose ecosystem designed for modern retail pharmacies:

For Pharmacy Operators (Backend Automation):
- Autonomous Inventory Manager: An agent that monitors stock levels, predicts demand spikes based on seasonal data, and drafts automated purchase orders.
- Workflow Coordinator: Coordinates tasks between staff, flag expiring batches, and optimizes daily fulfillment queues.
For Consumers (Personalized Wellness Guide):
- Wellness Concierge Agent: A conversational agent that understands user lifestyle goals, dietary restrictions, and existing health routines to recommend appropriate vitamins, supplements, and OTC wellness products.
- Smart Product Navigator: Guides users to the right retail products without overwhelming them with medical jargon.

🛠️ How we built it

We constructed PharmAssist AI using a modular, multi-agent architecture:

Agentic Framework: Utilized LangGraph / CrewAI to orchestrate tasks between specialized agents (Inventory Agent, Wellness Guide Agent, and Supervisor Agent).
Backend & LLM Orchestration: Built the core services using Python and FastAPI, integrating Claude 3.5 Sonnet / GPT-4o via LangChain for cognitive reasoning.
Database & Knowledge Base: Used PostgreSQL for transactional inventory data and Pinecone as a vector database to store and query complex product catalogs and wellness documentation.
Frontend: Developed a responsive, user-friendly dashboard using Next.js and Tailwind CSS for both the pharmacy backend view and the customer kiosk interface.

🚧 Challenges we faced

Agent State Management: Keeping the context synchronized between the Inventory Agent and the Workflow Coordinator when sudden data changes occurred was tough. We solved this by implementing a robust state-sharing layer in our agentic framework.
Guardrails & Accuracy: Wellness recommendations require extreme precision. We spent a significant amount of time building strict retrieval-augmented generation (RAG) guardrails to ensure the wellness agent never crosses into providing diagnostic medical advice, keeping recommendations purely focused on retail wellness products.

🎉 Accomplishments that we're proud of

Successfully designed a working multi-agent system where agents autonomously pass tasks to each other without human intervention.
Created a beautiful, intuitive interface that makes complex backend automation look incredibly simple for pharmacy staff.
Maintained low-latency agent responses, ensuring the customer-facing wellness guide feels like a fluid, real-time conversation.

📚 What we learned

We learned how to design efficient agent loops and avoid infinite logic traps when multiple agents collaborate.
Deepened our understanding of grounding LLMs using vector databases to ensure safe, highly relevant product recommendations in a retail environment.

🚀 What's next for PharmAssist AI

IoT Integration: Connecting our inventory agents directly to smart shelf sensors for real-time stock updates.
Voice Commerce: Upgrading the customer wellness guide to support natural voice interactions for accessible in-store kiosks.
Predictive Analytics: Incorporating advanced machine learning models to forecast localized wellness trends (e.g., predicting allergen product demands ahead of season change).