TraverseHealth

Vision

Healthcare navigation is broken for people who need it most. When you're sick, scared, or don't speak English fluently, figuring out which doctor to see, where they are, and whether your insurance covers them is an overwhelming second problem on top of already feeling terrible. TraverseHealth collapses that entire process—from raw symptom to appointment-ready provider—into a single, supportive conversation.

Technical Architecture

Built as a single-file full-stack JAC application on the Jaseci stack, the system utilizes a sequence of five specialized walkers:

Intake: Uses by llm() to extract symptoms from emotional, unstructured patient descriptions.
Triage: Classifies urgency using high-precision semstring design.
Routing: Navigates specialty requirements via clinical graph traversal.
Matching: Real-time reconciliation of the NPI Registry (legitimacy).
Prep: Generates multilingual visit instructions and summaries.

The "Messy Middle" Layer

By leveraging the Jaseci stack, we moved beyond basic prompt engineering to focus on pure application logic:

LLM Integration: Used by llm() to transform raw patient data into structured clinical decisions.

Critical Challenges & Safety

The Emergency Branch: Our highest priority was ensuring life-threatening symptoms (e.g., stroke or cardiac events) bypass standard routing. We engineered a zero-false-negative urgency classifier that immediately surfaces 911 and the nearest ER.

Data Reconciliation: We solved the "friction" of healthcare data using the free NPI Registry, providing a validated list of providers without a proprietary database.
Linguistic Barriers: By automating the preparation of multilingual instructions, we ensure that the navigation process remains accessible regardless of the user's primary language.