Every mother deserves Anchal.
India loses over 24,000 mothers every year to complications that are almost entirely preventable. The tragedy is not that these women never sought care — most of them did register for antenatal visits at their local Primary Health Centre. But somewhere between the first registration and the critical later visits, they silently dropped out. A young woman walks 18 kilometers to register in her third trimester, misses her next appointment during harvest season, and never returns. No one notices until it is too late.
ASHA workers — the incredible frontline health volunteers who serve rural India — carry the weight of this crisis on paper registers and memory. They have no way to know which of their 200+ registered women is about to disappear, no way to prioritize who needs a home visit today, and no tools to communicate in the woman's own language with the right cultural sensitivity. AnchalAI changes that.
AnchalAI is an end-to-end AI-powered maternal care intelligence platform that predicts which pregnant women are at risk of dropping out of antenatal care before they disappear — and equips ASHA workers with the exact words to bring them back.
At its core sits a Random Forest classifier trained on demographic, geographic, and behavioral features — age, distance from the nearest PHC, literacy, husband support, missed visits, harvest season, and more. The model scores every registered woman with a dropout risk percentage, and the dashboard surfaces them in priority order so the ASHA worker knows exactly who to visit first.
But prediction alone isn't enough. When an ASHA worker taps "Reach Out" on a high-risk woman, AnchalAI's multi-agent AI pipeline — built on Google's Agent Development Kit (ADK) — springs into action. The RiskAnalystAgent runs the ML model, the CommunicationAgent uses Gemini to craft a warm, culturally appropriate outreach message in Bengali, Hindi, or English, adjusting tone for age, literacy, and family dynamics. The EscalationAgent determines the right follow-up action: a routine check-in, an urgent home visit, or an immediate PHC alert.
The result is a complete Care Action Plan — risk score, human-readable risk factors, a vernacular message ready to send, and a clear escalation decision — delivered to the ASHA worker's dashboard in seconds.
While most existing maternal health solutions focus on basic digitization or reactive SMS reminders (sent after a missed visit), AnchalAI takes a fundamentally different approach:
- Predictive vs. Reactive: AnchalAI doesn't wait for a missed appointment. It predicts the likelihood of a dropout using an ML classifier before it happens.
- Hyper-Personalized & Context-Aware: Instead of robotic, generic texts, our Multi-Agent AI (powered by Gemini) crafts warm, culturally sensitive messages in Bengali, Hindi, or English, factoring in the mother's specific situation (e.g., distance, husband support).
- Zero-Cognitive-Load for ASHA Workers: The dashboard translates complex ML outputs into simple, actionable steps. A tap on "Reach Out" instantly generates a complete Care Action Plan.
"AnchalAI is the first maternal care platform designed for rural India that combines Predictive ML with Agentic Generative AI to turn raw health data into instant, culturally-sensitive interventions for frontline workers."
Phase 2: Scaling AnchalAI for a Billion Lives
- WhatsApp Integration for Zero-Friction Adoption: Pushing daily priority lists and localized outreach messages directly to the ASHA worker's personal WhatsApp, guaranteeing massive adoption with zero learning curve.
- OCR for Paper Registers: Enabling ASHA workers to snap a photo of their physical RCH registers. Gemini’s Multimodal Vision will extract handwriting securely into the AnchalAI database, eliminating manual data entry.
- Voice Bots & IVR: Integrating Voice LLMs to trigger automated, conversational phone calls in local dialects for mothers with limited text literacy.
- Offline-First Application: Expanding into an offline-first mobile app that seamlessly syncs when network connectivity returns in remote villages.
The dashboard surfaces the highest-risk women first. Each card shows the woman's age, distance from PHC, missed visits, and a bold risk percentage. The ASHA worker simply taps "Reach Out" to trigger the AI pipeline.
When "Reach Out" is tapped, AnchalAI shows why the woman was flagged (distance, missed visits, limited husband support), lets the ASHA worker choose a language, and generates a Gemini-powered outreach message in Bengali — warm, simple, culturally sensitive, and ready to send.
ASHA Worker → Dashboard → Flask API → ML Model (Random Forest)
↓
ADK Agent Pipeline
RiskAnalystAgent → CommunicationAgent → EscalationAgent
↓
Gemini (Vertex AI) → Bengali/Hindi/English Message
The system runs as two Cloud Run services:
- Backend API — Flask app serving the ML model and Gemini message generation for the dashboard
- ADK Agent — FastAPI app exposing the full multi-agent pipeline as a single
/predictendpoint
Both are containerized with Docker, deployed to Google Cloud Run (asia-south1), and the frontend is hosted on GitHub Pages.
| Layer | Technology |
|---|---|
| AI / LLM | Google Gemini 2.0 Flash, Google ADK (Agent Development Kit) |
| ML Model | scikit-learn Random Forest, pandas |
| Agent Orchestration | Google ADK SequentialAgent (RiskAnalyst → Communication → Escalation) |
| Backend | Flask (dashboard API), FastAPI (agent endpoint) |
| Infrastructure | Google Cloud Run, Vertex AI, Docker |
| Frontend | HTML/CSS/JS, GitHub Pages |
| Language | Python 3.11 |
| Service | URL |
|---|---|
| 🌐 Dashboard | mrmallick07.github.io/AnchalAI |
| ⚙️ Backend API | anchalai-backend-961197586142.asia-south1.run.app |
| 🤖 ADK Agent | anchalai-agent-961197586142.asia-south1.run.app |
Test the agent directly:
curl -X POST https://anchalai-agent-961197586142.asia-south1.run.app/predict \
-H "Content-Type: application/json" \
-d '{
"age": 19,
"distance_to_phc_km": 18.5,
"previous_pregnancies": 1,
"attended_last_visit": 0,
"household_income_level": 1,
"husband_support": 0,
"literacy": 0,
"trimester_at_registration": 3,
"harvest_season": 1,
"asha_visits_so_far": 1,
"language": "Bengali"
}'# Clone the repository
git clone https://github.com/mrmallick07/AnchalAI.git
cd AnchalAI
# Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate
# Install dependencies
pip install -r requirements.txt
# Add your Gemini API key
echo "GEMINI_API_KEY=your_api_key_here" > .env
# Run the Flask backend (dashboard API)
python app.py
# Or run the ADK agent endpoint
python -m agent.mainThe Flask app runs on http://localhost:8080. The agent endpoint also runs on port 8080.
To get a Gemini API key, visit aistudio.google.com/apikey.
Hannan Ali Mallick
📧 hannanmallick07@gmail.com 📞 +91 7003711453 🔗 linkedin.com/in/hannan-ali-mallick
Built with ❤️ for the mothers of India.