Health Equity Bridge (Severity)

Health Equity Bridge is an AI-assisted healthcare navigation app designed to reduce access barriers for patients who face language, cost, and care-navigation challenges.

It combines a Next.js frontend, a FastAPI backend, and a local multi-agent orchestration layer that can:

• translate and normalize symptom descriptions
• assess urgency with deterministic safety overrides
• recommend nearby care options
• estimate likely care costs
• generate provider-facing summaries
• support multilingual, stateful chat flows

Demo

Severity

Why This Project Exists

Many patients do not fail because care is unavailable. They fail because the path to care is confusing.

This project focuses on practical barriers:

• language mismatch between patients and providers
• uncertainty about urgency
• difficulty finding appropriate nearby care
• lack of cost transparency
• poor handoff quality when moving from chat to real care

What’s Implemented

Core user flow

• Conversational symptom intake with session state
• Structured follow-up questions when information is missing
• Triage with hybrid AI + deterministic emergency rules
• Care navigation using Google Places
• Cost estimation with fallback heuristics
• Provider-facing summary generation
• Streaming chat responses to the frontend

Extended workflows

• Visit note extraction from transcripts
• Conversation summarization
• Turn-by-turn translation
• Appointment workflow hooks, including VAPI-based scheduling integration
• User profile persistence via Firebase or local file fallback

Architecture

This is an agent-oriented monolith, not a distributed system.

Authoritative runtime path:

frontend -> backend/main.py -> backend/orchestrator.py -> a2a/router.py -> specialized/* -> tools/*

High-level execution flow

1. The frontend sends a symptom or chat request.
2. The backend creates a typed AgentMessage state object.
3. The orchestrator extracts semantic meaning and determines the next action.
4. Specialized agents handle language, triage, emergency guidance, navigation, cost, and provider handoff.
5. Tool adapters call Gemini, Google Places, Firebase, and scheduling services.
6. The backend returns structured outputs, session state, and trace data for UI visualization.

Specialized agents

• language_agent: language detection, simplification, translation
• triage_agent: urgency assessment with deterministic overrides
• emergency_agent: urgent guidance for high-risk symptom patterns
• navigation_agent: nearby care search
• cost_agent: care cost estimation
• contact_agent: provider handoff summary generation

Why It’s Technically Interesting

• Uses typed orchestration instead of free-form prompt chaining
• Combines LLM reasoning with deterministic safety logic
• Maintains multi-turn conversation state with follow-up resolution
• Streams responses and exposes traceable agent execution to the UI
• Includes fallback behavior when external AI or API dependencies fail

Tech Stack

Frontend

• Next.js App Router
• TypeScript
• Tailwind CSS

Backend

• FastAPI
• Pydantic
• Firebase / Firestore fallback storage

AI and orchestration

• Gemini
• Modular local multi-agent orchestration
• Local A2A-style agent handoff router

External integrations

• Google Places API
• VAPI

Repository Structure

frontend/        Next.js UI
backend/         FastAPI routes, orchestration entrypoints, persistence helpers
specialized/     Specialized agent implementations
tools/           Gemini, maps, cost, formatting, profile helpers
a2a/             Local A2A-style handoff router
core/            Base agent runtime, tracing, response assembly
shared/          Shared schemas and TS/Python type bridge
tests/           Orchestration and tool tests
docs/            Architecture notes and demo script