Inspiration
Our inspiration comes from a critical gap in clinical pharmacy: the "Snapshot Problem." Chronic conditions are often managed based on today’s lab values, ignoring the longitudinal velocity of change. Inspired by a family member's struggle with hyperthyroidism, we realized that AI's true "Endgame" in healthcare isn't just chatting about guidelines—it’s about interoperable intelligence. We built Anthronomic ACIF to bridge the gap between raw data bundles and life-saving clinical narratives.
What it does
Anthronomic ACIF (Agnostic Clinical Intelligence Framework) is an interoperable Clinical Operating System that manages the longitudinal journey of a patient.
- Autonomous Interoperability: Using SHARP context, the agent autonomously retrieves patient records from FHIR servers via a secure Go-based gateway.
- Longitudinal Intelligence: It computes the "Velocity of Change" (e.g., fT4 rate per week) to identify "Rapid Decline" risks—preventing iatrogenic hypothyroidism that snapshot monitoring misses.
- Evidence-Based Prognostics: It codifies validated medical models, like the GREAT Score, into a deterministic science engine.
- Actionable Deliverables: It synthesizes raw FHIR JSON and unstructured clinical notes into professional SBAR (Situation, Background, Assessment, Recommendation) reports
How we built it
We architected a decoupled, microservices-based IaaS:
- Go MCP Gateway: Built with the Model Context Protocol to handle high-concurrency SSE connections and SHARP context propagation.
- Python Science Engine: A FastAPI-based engine that executes deterministic clinical logic and mathematical modeling.
- Intelligence Fallback: Developed logic that allows the agent to extract data from unstructured .txt notes when structured FHIR feeds are fragmented.
- Infrastructure: Deployed on AWS EC2 using Docker, with ngrok providing secure TLS tunneling for the Prompt Opinion assembly.
Challenges we ran into
The "Last Mile" is paved with technical hurdles. We faced significant challenges with the MCP handshake, specifically injecting the ai.promptopinion/fhir-context extension into strictly-typed Go libraries. We solved this by developing a custom Stream Interceptor that live-edits SSE chunks. We also had to optimize our Science Engine to fit PyTorch-heavy dependencies within the 8GB constraints of base AWS volumes, eventually expanding the IaaS to 30GB to ensure production stability.
Accomplishments that we're proud of
- SHARP Integration: Achieving the "Green Check" on the Prompt Opinion platform, proving our server can securely handle X-FHIR-Access-Token and X-Patient-ID headers.
- Deterministic Math: Building a system that accurately calculates hormone velocity ($ \Delta fT4 / \Delta t $) to trigger urgent safety alerts.
- Agnostic Scaling: Proving that by simply swapping a Python router, the same infrastructure can manage Thyroid, Hypertension, or Glaucoma within a single A2A conversation.
What we learned
We learned that the power of AI in healthcare lies in A2A (Agent-to-Agent) collaboration. Seeing our "Specialist Agent" receive context from a "General Agent" and autonomously produce a structured SBAR report proved that interoperable standards like COIN and SHARP are the only way to scale clinical quality
What's next for Anthronomic Clinical OS: Interoperable Agnostic Intelligence
The next phase of Anthronomic is the Spatial Clinical Lab. We intend to:
- 1. Digital Twins: Map our longitudinal data onto 3D anatomical models for AR/VR visualization of disease states.
- 2. Expansion: Build out clinical routers for the 2025 ATA Oncology guidelines and ASCVD risk stratification.
- 3. EHR Native: Moving beyond the marketplace to direct integration with hospital Epic/Cerner environments.
Log in or sign up for Devpost to join the conversation.