MediSync

Inspiration

The inspiration for MediSync came from witnessing the overwhelming documentation burden faced by clinical professionals, who often spend over 16 hours a week on administrative tasks. We realized that while AI has made great strides, there was no open-source, developer-friendly SDK that allowed EHR startups and researchers to easily embed multi-agent clinical intelligence into their systems without the high costs and "black box" limitations of commercial alternatives.

What it does

MediSync is a multi-agent SDK that takes a FHIR R4 Bundle as input and coordinates three specialized AI agents to generate clinical insights. A Clinical Scribe drafts structured SOAP notes from encounter data; a Drug Interaction Analyst cross-references medications for safety warnings; and a Medical Coding Advisor suggests relevant ICD-10 and CPT codes. It provides a unified API, real-time streaming, and confidence-scored outputs to ensure clinical trust and developer speed.

How we built it

We built the MediSync website using React 19, Vite, and Framer Motion to create a high-performance, glassmorphic UI that reflects a futuristic medical aesthetic. The core logic is documented as a FastAPI backend orchestrating agents via LangGraph and LiteLLM. We focused heavily on FHIR R4 standards to ensure interoperability and used Vanilla CSS for a custom, premium design system without the overhead of heavy frameworks.

Challenges we ran into

One of the primary challenges was designing a system that could handle the complexity of medical data while remaining easy for developers to integrate. Orchestrating three agents in parallel required careful token budget management and state synchronization to produce a unified response in under 6 seconds. Additionally, creating a visual design that felt "medical" yet "cutting-edge" required iterating on custom glassmorphism effects and complex CSS animations.

Accomplishments that we're proud of

We are incredibly proud of building a comprehensive 25-page developer platform that includes a live functional playground, a visual FHIR bundle builder, and deep technical documentation. Successfully integrating a multi-agent orchestration pattern that remains standard-compliant (FHIR) while being open-source is a major milestone for us in the clinical AI space.

What we learned

Through this project, we deepened our understanding of the FHIR R4 ecosystem and the nuances of agentic orchestration in high-stakes environments like healthcare. We learned that the "Developer Experience" (DX) is just as important as the AI's accuracy; having searchable SDK references and interactive debugging tools (like our Output Inspector) is what makes a project actually usable in production.

What's next for MediSync

Our next steps involve expanding the agent pool to include Radiology Report Analysis and Clinical Trial Matching. We are also planning to release a Helm chart for simplified enterprise deployment and a local-first "Library Mode" that allows MediSync to run entirely air-gapped on clinical workstations using local models like Llama 3.1.

Built With

• anthropic-claude-3.5-sonnet
• biobert
• css3-(custom-design-system)-frameworks/libraries:-react-19
• drugbank
• framer-motion-(animations)
• html5
• javascript-(es6+)
• langchain
• langgraph-(agent-orchestration)
• litellm-(model-gateway)
• llamaindex-(rag)-ai-models:-openai-gpt-4o
• lucide-react-(icons)
• ollama-(llama-3.1)-databases:-postgresql-16-+-pgvector-medical-standards:-fhir-r4
• react-router-v7-deployment-&-cloud:-vercel-core-sdk-&-agent-architecture-(documented)-languages:-python-3.11+-frameworks:-fastapi
• rxnorm
• vite-6