WhisperCare

India-focused, voice-first healthcare workflow platform.

WhisperCare is a FastAPI + LangGraph + LiveKit backend for patient onboarding, consent, voice-led intake, triage, clinician handoff, care planning, and ABDM/FHIR-ready exports. It is designed around Indian telehealth workflows, multilingual expansion, and config-driven deployment.

WhisperCare is a healthcare workflow assistant. It is not a medical device, does not diagnose, does not prescribe, and does not replace a Registered Medical Practitioner.

What It Does

• Patient onboarding with phone-first profile fields, emergency contact, language, and optional ABHA identifiers.
• Consent capture for telehealth and AI-assisted workflows with audit logging.
• Voice-first intake using LiveKit session provisioning and a worker bridge.
• Mental-health screening with PHQ-9 and GAD-7 style scoring from YAML configuration.
• India safety routing for urgent symptoms and crisis signals.
• Clinician handoff through intake summaries, appointments, visits, and care-plan foundations.
• ABDM/FHIR-ready export preview using FHIR-shaped Patient and QuestionnaireResponse resources.
• Dockerized local stack with API, Postgres, LiveKit, and voice-worker services.

Architecture

Browser / Mobile
      |
      | REST + LiveKit token
      v
FastAPI Backend -------------- PostgreSQL
      |                             |
      | LangGraph                   | audit, patients,
      v                             | intakes, visits
Intake / Triage Graph
      ^
      |
LiveKit Voice Worker <-------> LiveKit Server
      |
      | STT / TTS / LLM providers configured by YAML
      v
Patient voice session

Core services:

• api: FastAPI application.
• postgres: primary relational datastore.
• livekit: local LiveKit SFU for voice rooms.
• voice-worker: LiveKit/backend bridge. Current mode is a safe stub that stays alive and exposes configured runtime details.

Tech Stack

Area	Choice
API	FastAPI
Workflow orchestration	LangGraph
Voice transport	LiveKit
Database	PostgreSQL via async SQLAlchemy
Migrations	Alembic
Config	YAML + env-var references
Tests	Pytest
Container	Docker / Docker Compose

Quick Start

Clone and run the full stack:

git clone https://github.com/SarthakShrivastav-a/WhisperCare.git
cd WhisperCare
copy .env.example .env
docker compose up --build

Open:

• API health: http://localhost:8000/api/health
• Demo page: web/index.html
• LiveKit HTTP port: http://localhost:7880

Run locally without Docker:

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Run tests:

python -m pytest -q

Docker Image

The GitHub Actions workflow publishes Docker images to Docker Hub after merged pull requests.

Current image namespace:

sarthak73/whispercare

Expected release tags:

sarthak73/whispercare:latest
sarthak73/whispercare:main-<short_sha>
sarthak73/whispercare:dev-latest
sarthak73/whispercare:dev-<short_sha>

Required GitHub Actions secrets:

DOCKERHUB_USERNAME
DOCKERHUB_TOKEN

Configuration Policy

All runtime values live in config/app.yaml.

Do not hardcode:

• model names
• provider names
• LiveKit agent names
• LiveKit URLs
• emergency numbers
• crisis response text
• scoring thresholds
• supported languages
• Docker image repositories
• API base URLs

Secrets are referenced by environment variable names in YAML, then resolved at runtime.

Important config files:

config/app.yaml
config/livekit.dev.yaml
.env.example
docker-compose.yml

API Overview

All routes are mounted under /api.

Method	Route	Purpose
GET	/health	Service health check
POST	/patients	Create patient profile
GET	/patients/{patient_id}	Read patient profile
PATCH	/patients/{patient_id}	Update patient profile
POST	/consents	Capture patient consent
GET	/consents/{patient_id}	List patient consents
POST	/intake/sessions	Start intake session
PUT	/intake/sessions/{session_id}	Advance intake graph
GET	/intake/sessions/{session_id}	Read intake session
POST	/voice/connect	Create LiveKit voice session token
POST	/voice/disconnect	Close voice session
POST	/appointments	Create appointment
POST	/visits/summary	Create visit summary
GET	/abdm/export/{patient_id}	Export FHIR-shaped bundle

India Safety Defaults

Configured emergency and crisis resources:

• National emergency: 112
• Ambulance defaults: 108, 102
• Tele-MANAS: 14416, 1-800-891-4416
• KIRAN: 1800-599-0019

Safety behavior:

• Crisis signals stop the regular intake path.
• Urgent cases are flagged for human review.
• AI responses include escalation guidance.
• The system prepares clinical context; it does not provide diagnosis or prescriptions.

Project Layout

app/
  api/              FastAPI routers
  core/             YAML config loader
  db/               async SQLAlchemy setup
  graphs/intake/    LangGraph intake flow
  livekit/          voice worker bridge
  models/           SQLAlchemy models
  schemas/          Pydantic schemas
  services/         business logic
config/
  app.yaml
  livekit.dev.yaml
docs/
tests/
web/

Development Workflow

Branch model:

• main: release branch.
• dev: integration branch.
• feature/*: scoped work branches.

Preferred flow:

git checkout dev
git pull origin dev
git checkout -b feature/my-change

# work, test, commit
python -m pytest -q

git push -u origin feature/my-change
gh pr create --base dev --head feature/my-change
gh pr merge --merge

Release flow:

gh pr create --base main --head dev --title "release: merge dev into main"
gh pr merge --merge

Current Status

Implemented foundation:

• FastAPI backend shell
• YAML config loader
• async DB setup
• patient and consent modules
• intake graph
• PHQ-9/GAD-7 scoring
• India safety routing
• LiveKit connect/disconnect foundation
• Docker Compose stack with LiveKit
• ABDM/FHIR-style export preview

Next meaningful work:

• Replace voice-worker stub with full LiveKit AgentSession loop.
• Add real OTP/auth flow.
• Add Alembic generated migration revisions.
• Expand clinician dashboard APIs.
• Add production ABDM sandbox integration.
• Add provider-specific STT/TTS/LLM adapters.

License

MIT. See LICENSE.