RuralHealth AI

RuralHealth AI is an offline-first community healthcare screening and referral decision-support platform designed to support ASHA workers, ANMs, and Primary Health Centre (PHC) teams in rural and semi-urban communities.

The platform addresses a practical challenge in community healthcare: frontline workers often need to collect patient information, perform preliminary screening, identify cases requiring further attention, coordinate referrals, and maintain follow-up records while working with limited connectivity and resources.

RuralHealth AI brings these activities into a unified digital workflow.

Core Capabilities

The platform provides a structured patient screening workflow covering patient information, symptoms, vital measurements, risk factors, review, screening results, and recommended next steps. It supports explicit handling of unavailable measurements and incomplete information rather than assuming missing values.

A rule-based clinical risk management layer classifies screening outcomes into states such as Low, Moderate, High, Needs Review, and Insufficient Data. The system is designed so that clinical rules remain authoritative and the AI assistant cannot modify or override the official risk classification.

The referral and appointment modules allow healthcare workers to create referrals, track referral status, coordinate appointments, and manage follow-up activities. The platform also provides a healthcare facility discovery workflow using Google Maps and location services for identifying nearby hospitals, PHCs, clinics, and other relevant facilities.

Offline-First Architecture

RuralHealth AI is designed for environments where reliable internet connectivity cannot be assumed. Core application data can be maintained locally and synchronized when connectivity becomes available.

The frontend is implemented as a Progressive Web Application with local browser storage, while the backend uses FastAPI and a structured database layer. Synchronization states are explicitly represented so that locally created records can be tracked through queued, syncing, synced, failed, or conflict states.

Local AI Assistant

The platform includes a locally hosted AI assistant using Ollama and Gemma 3 270M. The assistant is intended for health education, workflow guidance, multilingual communication, summarization, and interpretation of approved knowledge and health-data resources.

The assistant supports English, Hindi, and Bengali and is designed to operate through the local AI runtime rather than sending user conversations directly to an external AI service.

The AI assistant is deliberately separated from the clinical decision engine. It does not independently diagnose patients, prescribe medication, generate official risk scores, or override clinical rules. Where information is insufficient, the system is designed to indicate that further review is required.

Health Data Intelligence

RuralHealth AI also incorporates a Kolkata health dataset derived from HMIS and NFHS-5 data. The current processed dataset contains 14 records and 1,957 features, with 8,202 HMIS numeric records processed and 73 NFHS-5 features extracted. HMIS provides the yearly data backbone, while the NFHS-5 values correspond to the 2019–20 survey period.

Rather than using the structured dataset as direct chatbot training data, the platform uses it as a data-intelligence layer. Users can query available health indicators, explore relevant features, retrieve supported historical values, and obtain dataset-based summaries through the assistant.

Privacy and Safety

The system follows a human-in-the-loop approach. Patient records, aggregate health datasets, clinical rules, and AI-generated explanations are treated as separate layers.

The AI assistant provides supporting information; healthcare professionals and approved clinical workflows remain responsible for clinical decisions. The platform is designed with role-based access, consent, auditability, controlled data exposure, and privacy-conscious local processing in mind.

Technology Stack

Frontend: React, TypeScript, Vite, Tailwind CSS, Progressive Web App

Backend: Python, FastAPI, REST APIs

Data: SQLite, IndexedDB/Dexie, Pandas, NumPy

AI: Ollama, Gemma 3 270M, Retrieval-Augmented Generation

Healthcare Intelligence: Rule-based risk assessment, machine-learning support, HMIS and NFHS-5 datasets

Location Services: Google Maps Platform, Google Places API, Browser Geolocation

Objective

The objective of RuralHealth AI is not to replace healthcare professionals, but to strengthen the workflow around them—making community-level screening, risk review, referral coordination, follow-up, and health information more accessible, structured, and resilient in low-connectivity environments.