Health Nav: Al-Powered Healthcare Triage & Navigation Agent

Inspiration

We were inspired by the often stressful and confusing process of seeking medical care. Patients frequently face long wait times, unnecessary ER visits for non-emergencies, and difficulty determining the appropriate care level for their symptoms. The current system is inefficient, leading to patient frustration and wasted healthcare resources. We envisioned a world where patients could get immediate, accurate guidance and seamless access to care—making healthcare more accessible, efficient, and patient-centric.

What it does

Health Nav is an agentic AI system that functions as a real-time health navigator and care coordinator. It automatically triages patient symptoms, guides them to the right care (clinic, ER, or tele-consult), and handles the entire logistical process. Symptom Triage & Entity Extraction: It takes natural language symptom descriptions and analyzes them for medical urgency. Intelligent Care Routing: It uses the triage result to determine the appropriate next step. End-to-End Automation: It executes complex administrative workflows like: Appointment Booking & Insurance Verification. Sending automated follow-up reminders. Providing a secure, personalized patient portal.

How we built it

Health Nav was built as a robust, scalable, and highly secure platform leveraging a wide array of specialized AWS services. Agentic Core (AI/ML): The core triage logic utilizes Amazon Bedrock for access to powerful Foundation Models (FMs) that perform complex reasoning and conversation, and we used Amazon SageMaker for custom model development and deployment of supporting machine learning components. Crucially, we use Amazon Comprehend Medical to accurately extract structured medical entities (symptoms, medications, conditions) from the patient's unstructured text input, which feeds into the triage model. Workflow Automation: The "agentic" capability is orchestrated using AWS Step Functions. This service manages the complex, multi-step workflow—from triage result, to insurance check, to booking—and provides built-in fault tolerance. The actual business logic for booking APIs and sending notifications is executed using AWS Lambda functions, all exposed and managed via Amazon API Gateway. Data & Security: Patient data is stored securely and scalably in Amazon DynamoDB. Large assets and documents are kept in Amazon S3. User authentication and authorization are managed by Amazon Cognito. Monitoring & Analytics: Amazon CloudWatch provides comprehensive monitoring and logging for the entire system. We utilized Amazon OpenSearch for logging and powerful search capabilities across patient records and engagement history, and Amazon QuickSight for business intelligence dashboards to visualize triage accuracy and system performance. Front-End: The patient-facing mobile/web application was deployed and managed using AWS Amplify.

Challenges we ran into

The primary challenges were centered around safety, compliance, and integrating complex workflows: Ensuring Clinical Safety: Combining the generative nature of models in Bedrock with the strict, analytical extraction of Comprehend Medical required careful engineering to ensure the triage guidance was always safe and medically responsible. Orchestration Complexity: The booking and verification process involves multiple external systems. We utilized AWS Step Functions to manage these stateful, long-running tasks, which helped simplify error handling and retries, but required careful design of the underlying Lambda functions. Data Security and HIPAA Compliance: Designing the system to be HIPAA-eligible required rigorous implementation of security controls across all services, ensuring patient data remained secure in transit and at rest (DynamoDB, S3).

Accomplishments that we're proud of

We are incredibly proud of Health Nav's demonstrated efficiency improvements and its robust architecture. Agentic Efficiency: By using Step Functions to automate complex admin tasks, we successfully demonstrated the ability to automate 70% of the administrative pre-care process, saving both patients and providers valuable time. Medical Accuracy: Our internal testing showed a 95% correlation with human clinician-led triage decisions for common conditions, achieved by leveraging the domain expertise of Comprehend Medical and the reasoning power of Bedrock. Scalability & Security: The architecture built entirely on managed, HIPAA-eligible AWS services is inherently secure and globally scalable, ready to serve millions of patients across different regions.

What we learned

This project reinforced the critical importance of selecting the right tool for the right job within a large cloud ecosystem. We learned how to successfully combine the power of modern Generative AI (Bedrock) with domain-specific ML tools (Comprehend Medical) to create a high-stakes, accurate application. Furthermore, mastering Step Functions for workflow orchestration was key, demonstrating that true agentic AI systems are more about coordinating actions and managing state than just generating text.

What's next for Health Nav: Al-Powered Healthcare Triage & Navigation Agent

Phase 1: Deep Integration & Workflow Automation The immediate focus is eliminating information silos and automating the patient’s administrative journey. EHR/EMR Bi-directional Integration: We will develop real-time, two-way connections using AWS Lambda and API Gateway to integrate with major Electronic Health Record (EHR) systems (FHIR/HL7 standards). This allows Health Nav to securely write the Triage Summary Report directly into the patient's chart and retrieve historical data for richer context. Context-Rich Triage: The system's agentic capabilities will be enhanced beyond simple text analysis by integrating real-time data from wearables and external health apps. We will use Amazon Kinesis/EventBridge to feed this event data into the Amazon Bedrock models, allowing for a personalized risk assessment (e.g., correlating symptoms with recent sleep or activity patterns). Provider Handoff Automation: We will leverage Amazon Comprehend Medical and Bedrock to automatically generate a concise, structured "Consultation Prep Summary" for the receiving clinician, drastically reducing pre-visit administrative time. Phase 2: Clinical Expansion & Predictive Intelligence The next phase scales the clinical impact and introduces proactive care. Specialty Domain Expertise: We will expand the model's knowledge base to cover specialized medical domains like Behavioral Health, Chronic Disease Management, and Pediatrics. This unlocks new, complex patient populations and increases the system's overall utility. Predictive Health Modeling: Utilizing aggregated data securely stored in DynamoDB and analyzing it with Amazon SageMaker, we will develop predictive models. These models will forecast critical outcomes, such as the likelihood of hospital readmission or the need for a follow-up test, enabling proactive, automated patient outreach and care coordination. Global Readiness: Full multilingual support will be implemented using Bedrock's multilingual Foundation Models to prepare the platform for global deployment, ensuring high accuracy and regulatory adherence across different regions. Phase 3: Commercialization and Scaling The final phase centers on validation and building a sustainable business model. Clinical Pilot Programs: We will partner with large healthcare systems for structured pilot programs to validate the Return on Investment (ROI), specifically measuring the reduction in unnecessary ER visits and administrative costs. Tiered Monetization: The business model will transition to a Subscription-as-a-Service (SaaS) structure, potentially adopting a usage-based or outcome-based pricing approach that rewards Health Nav for the efficiency and cost savings it delivers to partner hospitals. The foundation on AWS allows for seamless scaling to meet the demands of any size healthcare organization globally.