SanRaksha is an open‑source Mission for Optimal Motherhood, AI‑driven platform that helps ASHA workers, ANMs, and PHC/CHC staff detect and track high‑risk pregnancies in rural India.
It works fully offline on low‑cost Android phones and syncs when connectivity returns — ensuring healthcare continuity even in no-network zones. Loading with a dashboard allowing PHC staff to track block-wise risk and complaints
| Category | Feature | Why It Matters |
|---|---|---|
| Offline‑First | Risk-scoring runs on device, no internet needed | ASHA workers often work in low-connectivity regions |
| Hybrid Models | Offline: Compact neural net Online: XGBoost |
Combines speed + accuracy + explainability |
| Layered Risk Logic | Two logistic-regression layers before the neural net | Transparent factor-wise scoring for ASHA workers |
| Modular & Extensible | FL-ready via Flower templates | Future-proof, privacy-preserving updates |
| NLP Data Entry | Extracts vitals like BP, HR, Sugar from ASHA voice notes | Speeds up data entry for low-literacy field workers |
| PHC Dashboard | Streamlit-based geolocation dashboard with map, pie charts, reports, and analytics | Allows PHC staff to track block-wise risk and complaints |
| Feature Abbr. | Full Name & Unit | Typical Range |
|---|---|---|
| BMI | Body Mass Index (kg/m²) | 16 – 40 |
| BS | Random Blood Sugar (mmol/L) | 4–15 |
| HR | Heart Rate (beats/min) | 60–140 |
| BT | Body Temperature (°F) | 96–100 |
| PrevComp | Previous Pregnancy Complications (binary) | 0 / 1 |
| PreDM | Pre‑existing Diabetes (binary) | 0 / 1 |
| GDM | Gestational Diabetes (binary) | 0 / 1 |
| MentHlth | Mental‑Health Concerns (binary) | 0 / 1 |
LR‑A (Obstetric History Layer)
Inputs: PrevComp, PreDM, GDM, MentHlth → Output: Score A
LR‑B (Vitals Layer)
Inputs: HR, BT, BS, BMI → Output: Score B
LR‑C (Meta Layer)
Inputs: Score A, Score B → Output: Final Risk Score (0–1)
Offline Neural Net
Input(2)→ Dense(16, ReLU) → Dense(8, ReLU) → Dense(1, Sigmoid)- Refines Score using non-linear field data patterns
Online Model (XGBoost)
- Same inputs, synced when internet returns
- Used for dashboard analytics & deeper predictions
- ASHA workers speak vitals ("BP 130/90, sugar 120")
- A lightweight Whisper model (converted via OpenAI Whisper) runs locally
- A regex-based parser extracts and autofills:
- BP, Sugar, HR, Temp, BMI
- Designed for fully offline Android execution with Whisper inference
Powered by OpenAI Whisper – an automatic speech recognition system for multilingual voice-to-text transcription.
Built with Streamlit, the dashboard provides:
- Geolocation map of risk cases (Folium + Plotly)
- Risk pie charts, bar graphs, daily checkup stats
- Complaint panel (via form + auto CSV save)
- Downloadable reports
- India-focused view with filtering by state
Perfect for PHC/CHC staff to monitor local maternal risk trends.
- Early referrals for rural pregnant women
- ASHA/ANM workers receive real-time alerts
- PHC/CHC staff track block-wise and daily trends
- NGOs and researchers access open-data reports
| Name | Role |
|---|---|
| Arindol Sarkar | ML Pipeline + Risk Scoring Models |
| Atul Gadkoti | Android App + Offline Sync |
| Ishita Singh | Web Dashboard + Geolocation Visualization + NLP Integration |
We welcome collaborations in clinical validation, federated learning, and dataset curation.
| Layer | Technology |
|---|---|
| App | Kotlin + TFLite |
| Server | FastAPI |
| ML | TensorFlow, scikit-learn, XGBoost |
| Dashboard | Streamlit, Plotly, Folium |
| FL-ready | Flower (client + server templates, in progress) |
Apache License 2.0 — see LICENSE file
Attribution details available in NOTICE
# Clone the repository
git clone https://github.com/sys6-exe/SanRaksha
cd sanraksha
# Run backend API server
cd server && uvicorn main:app --reload
# Run NLP + ML models (Jupyter)
cd ml_models && jupyter notebook
# Launch Dashboard
streamlit run dashboard/app.pyOpen an issue, start a discussion, or email 24cd3007@rgipt.ac.in We’re especially keen on:
-
Clinical validation partnerships
-
Rural deployment pilots (PHC/CHC, NGOs)
-
Dataset sharing under open‑data agreements
Let’s make maternal healthcare safer and more accessible. 🚑