SenseAI — AI-Powered Rural Healthcare Assistant

Bringing intelligent, real-time clinical guidance to community health workers — even without internet.

Inspiration

Access to timely healthcare guidance remains a major challenge in rural and low-resource areas. Community Health Workers (CHWs) often have to make critical decisions with limited tools, delayed diagnostics, and poor connectivity.

We were inspired to build a solution that could assist them with real-time risk assessment and actionable guidance — even without internet access.

What It Does

SenseAI is an AI-powered rural healthcare assistant that helps assess patient conditions and provide structured medical guidance. It supports multiple conditions including:

Maternal Hemorrhage
Tuberculosis Adherence
Pesticide Exposure
Diabetic Foot Ulcers (DFU)

The system collects patient inputs, processes them through specialized AI agents, and outputs:

Output	Description
📊 Risk Score & Level	Quantified severity assessment
📝 Clear Explanation	Human-readable reasoning
✅ Actionable Checklist	Step-by-step guidance for CHWs
⚠️ Missing Data Indicators	Flags incomplete assessments

It works across multiple platforms:

📱 Mobile App
💬 WhatsApp Chatbot
📞 Voice Call Agent

Key Features

1. Multi-Channel Access

Flutter mobile application for CHWs
WhatsApp chatbot for easy accessibility
Voice AI agent for hands-free interaction

2. AI-Based Diagnosis System

Domain-specific agents for Maternal Hemorrhage, TB Adherence, Pesticide Exposure, and DFU
Structured scoring-based risk assessment (not random AI output)

3. Offline Functionality

Uses lightweight Gemma model running locally on device
Provides basic reasoning when internet is unavailable
Automatically syncs data when connectivity is restored

4. Image-Based Analysis

DFU detection using image input
Evaluates wound severity and infection risk

5. Actionable Outputs

Risk level classification: LOW, MEDIUM, HIGH, CRITICAL
Step-by-step checklist for CHWs
Clear explanation in simple language
Missing data identification

6. Patient Management

Store patient records locally
Track diagnosis history
Save checklist progress

How We Built It

SenseAI uses a hybrid AI architecture designed for real-world, low-resource environments.

Frontend

Flutter mobile application for Community Health Workers

Backend

Flask APIs handling different diagnosis agents

AI Models

Model	Role
DeepSeek	Reasoning and structured decision-making
VisionPro + Custom Segmentation	DFU image analysis
Gemma 2B (quantized)	Offline inference via `llama.cpp`

Integrations

💬 WhatsApp Cloud API — Chat-based interaction
🎙️ ElevenLabs — Voice-based AI agent

Each condition is handled by a dedicated agent with structured scoring logic to ensure reliable and consistent outputs.

Challenges We Ran Into

Running AI models offline on low-end devices due to RAM and storage limitations
Managing real-time switching between online and offline modes
Ensuring AI outputs were not random by enforcing structured scoring logic
Integrating multiple platforms (Flutter, WhatsApp, Voice) into a single workflow
Handling image-based diagnosis reliably with limited datasets

🏆 Accomplishments We're Proud Of

✅ Built a fully functional hybrid AI system (online + offline fallback)
✅ Developed multi-channel access (mobile app, WhatsApp, voice calls)
✅ Implemented clinical-style scoring instead of generic AI responses
✅ Successfully integrated image-based DFU analysis
✅ Created a scalable solution tailored for rural healthcare challenges

📚 What We Learned

"Prompt engineering alone is not enough — structured reasoning is critical."

Offline AI deployment requires strong optimization and trade-offs
Simplicity in UI/UX is essential for real-world adoption
Multi-platform systems need strong coordination and data consistency
Building for real-world constraints is very different from building demos

What's Next for SenseAI

[ ] Optimize offline models to run on low-end devices
[ ] Expand support to more diseases — malaria, anemia, child health
[ ] Add real-time doctor consultation and telemedicine integration
[ ] Improve image analysis with better datasets and model training
[ ] Support more regional languages and voice interactions
[ ] Reduce dependency on paid APIs using open-source alternatives

Tech Stack

Frontend       → Flutter (Mobile App)
Backend        → Flask (Python)
AI (Online)    → DeepSeek, VisionPro
AI (Offline)   → Gemma 2B via llama.cpp
Messaging      → WhatsApp Cloud API
Voice          → ElevenLabs
Image Analysis → Custom Segmentation Model