Sahayi is a voice-first health-companion prototype built around phone calls and Indic-language speech. Its backend connects telephony, speech recognition and synthesis, conversation orchestration, retrieval, and a browser dashboard.
flowchart LR
A[Twilio call] --> B[Speech input]
B --> C[Signal and safety checks]
C --> D[Conversation agents and retrieval]
D --> E[Speech response]
C --> F[Escalation path]
D --> G[Dashboard and session record]
| Area | Tools |
|---|---|
| Backend | Python, FastAPI, SQLAlchemy, SQLite, WebSockets |
| Telephony | Twilio Voice |
| Speech | Sarvam AI integrations |
| Models | Google Gemini and configured model clients |
| Retrieval | ChromaDB, sentence-transformers, PubMed E-utilities |
| Dashboard | React, Vite, Firebase Auth |
Backend:
git clone https://github.com/ChristopherJoshy/Sahayi.git
cd sahayi/sahayi/backend
python -m venv .venv
# Windows: .venv\Scripts\Activate.ps1
# macOS/Linux: source .venv/bin/activate
pip install -r requirements.txt
Copy-Item .env.example .env # PowerShell; use cp on macOS/Linux
uvicorn main:app --reloadFrontend, in another terminal:
cd sahayi/sahayi/frontend
npm install
npm run devConfigure Twilio, Sarvam, model, database, and Firebase values locally. Do not put provider credentials in source control.
sahayi/backend/ API, call handling, agents, retrieval and safety logic
sahayi/frontend/ dashboard and session views
sahayi_flow.svg high-level call flow
MIT. See LICENSE.
This prototype is not a diagnostic or emergency-response service. It has not been clinically validated. Call recordings and health details are sensitive; use synthetic data for development and require qualified review before any real-world deployment.