MediSense AI — Smart Health & Supply Chain Resilience

Track: Resilience — Hack2Skill "Code for Communities 2"

An AI-powered early-warning platform that predicts medicine stock-outs before they happen and recommends cross-region redistribution to prevent them. It turns fragmented health-centre data into one shared view of risk — tracking stock, bed capacity, patient footfall, and staff attendance across India's Primary Health Centres so shortages are caught before they become crises.

Live demo: https://smart-health-brics.vercel.app

Why Resilience

MediSense AI strengthens the resilience of India's primary healthcare network by giving PHC administrators real-time visibility into the four operational signals that most often fail silently until they become a crisis: medicine stock, bed capacity, patient footfall, and staff attendance. AI-generated alerts and redistribution suggestions let administrators act before a stock-out or bed shortage disrupts care, and Hindi-language output ensures the tool is usable by staff across language backgrounds, not just English-fluent administrators.

Demo Access (for judges)


Email	`demo@smarthealth.com`
Password	`judge-demo-2026`

The demo account is read-only. You can explore the dashboard, alerts, risk scores and redistribution suggestions. The admin pages (add PHC, stock, attendance) are restricted to the admin role and are hidden from this account on purpose, so the shared demo data stays intact.

The Problem

Public health centres (PHCs) across India's states and districts often can't see medicine shortages, bed shortages, or staffing gaps coming until it is too late. Nobody can easily tell that one region has surplus stock while a neighbouring region is critically short — and this is true at national scale, not just in a single city or state.

What It Does

Tracks medicine stock, daily consumption, bed capacity/occupancy, patient footfall and staff attendance across PHCs in every Indian state and district
Predicts how many days of stock remain for every medicine
Generates plain-language early-warning alerts with Gemini, in English or Hindi
Matches surplus stock in one state against shortages in another and recommends a transfer
Combines stock and attendance into a risk score per PHC, with an AI-written explanation
Includes a Gemini-powered chat assistant that answers admin questions using live stock and redistribution data
Lets admins add and remove PHCs, stock, bed counts, and attendance records through simple forms

Real / Realistic Data

Live PHC inventory and footfall APIs aren't publicly available in India, so this project uses:

Real Indian geography — every state and district, not a hardcoded single city
Simulated stock depletion — a background job models realistic daily medicine consumption over time, so the dashboard reflects real depletion dynamics rather than static numbers

How It Works

┌─────────────┐     ┌──────────────────┐     ┌──────────────────────┐
│  React app  │────▶│  Express API     │────▶│  MongoDB Atlas       │
│  (Vercel)   │ JWT │  (Render)        │     │  PHCs, stock, users  │
└─────────────┘     └───────┬──────────┘     └──────────────────────┘
                            │
              ┌─────────────┴─────────────┐
              ▼                           ▼
     ┌─────────────────┐        ┌──────────────────┐
     │ Gemini API      │        │ FastAPI service  │
     │ alert, redist,  │        │ risk score       │
     │ risk explain,   │        │ (Render)         │
     │ chat — EN/HI    │        │                  │
     └─────────────────┘        └──────────────────┘

Stock, bed, footfall and attendance data lives in MongoDB.
The API calculates days of stock remaining for each medicine.
Gemini turns that number into a short alert for the dashboard, in English or Hindi based on the user's language toggle.
The API compares stock across PHCs to find surplus-versus-shortage matches, and Gemini writes the recommendation.
The Python service combines stock risk (65%) and staffing risk (35%) into a 0–100 score per PHC. Gemini then explains it. If the Python service is unreachable, the API computes the same score locally.
A Gemini-powered chat assistant answers free-form admin questions using the current alerts and redistribution data as context.
The dashboard shows everything live and refreshes automatically.

Google AI Studio & Gemini

All four Gemini prompts below were designed and iterated on in Google AI Studio using gemini-3.8-flash, then moved into the codebase. Production runs on gemini-3.5-flash-lite for its higher free-tier throughput; the same prompts and behavior carry over unchanged.

Alert prompt:

Write a short, urgent one-sentence alert under 25 words for a hospital admin dashboard. {medicine} stock at {phc} in {state}, India will run out in {days} days. Be direct and actionable. Output only plain text, no JSON, markdown, or bullet points. Respond in {English/Hindi}.

Redistribution prompt:

Write a short, actionable one-sentence recommendation (under 30 words) for a hospital admin dashboard. Suggest transferring {amount} units of {medicine} from {fromPhc} in {fromState} to {toPhc} in {toState}, which is critically low. Be direct. Output only plain text, no JSON, markdown, or bullet points. Respond in {English/Hindi}.

Risk explanation prompt:

Write one short sentence (under 25 words) explaining why {phc} in {state}, India has a risk score of {score}/100. Lowest stock has {days} days remaining. Staff attendance is {rate}%. Be direct and specific. Output only plain text, no JSON, markdown, or bullet points. Respond in {English/Hindi}.

Chat assistant prompt:

You are an assistant embedded in a hospital supply-chain dashboard called MediSense AI. Answer the admin's question using ONLY the data below. Be concise (under 80 words), direct, and specific. If the data doesn't cover the question, say so plainly. Respond in {English/Hindi}.

Tech Stack

Layer	Technology
Frontend	React 19, TypeScript, Vite, Tailwind CSS, TanStack Query, React Router
Backend	Node.js, Express 5, TypeScript, MongoDB, Mongoose, JWT
AI	Google Gemini — `gemini-3.8-flash` for prompt design in AI Studio, `gemini-3.5-flash-lite` in production. Powers alerts, redistribution, risk explanation, and chat, in English or Hindi
Risk scoring	Python, FastAPI
Hosting	Vercel (frontend), Render (API and risk service), MongoDB Atlas (database)

Project Structure

├── client/           React + Vite frontend
│   └── src/          pages, components, auth context, API helpers
├── server/           Express API
│   └── src/
│       ├── routes/       auth, phc, stock, attendance, country, risk, chat
│       ├── models/       Mongoose schemas
│       ├── middleware/   JWT auth, role checks, rate limiting
│       └── utils/        forecasting, redistribution, Gemini helpers
└── python-service/   FastAPI risk-score service

Getting Started

Prerequisites

Node.js 20+
Python 3.10+
A MongoDB database (local or Atlas)
A Gemini API key

1. Clone

git clone https://github.com/Parthwebde12/MediSense-AI.git
cd MediSense-AI

2. Risk service (optional but recommended)

cd python-service
python -m venv venv
venv\Scripts\activate        # macOS/Linux: source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --port 8001

If you skip this, the API falls back to computing the same score itself.

3. Server

cd server
npm install

Create server/.env:

PORT=5000
MONGO_URI=your_mongodb_connection_string
JWT_SECRET=a_long_random_string
GEMINI_API_KEY=your_gemini_api_key
PYTHON_SERVICE_URL=http://localhost:8001
SIMULATE_CONSUMPTION=false

Create the country record and your first admin account, then start the server:

npm run seed
npm run create-user -- --name "Your Name" --email you@example.com --password "a-strong-password" --role regional_admin
npm run dev

npm run seed deletes all PHCs, stock and attendance records before re-creating countries. Run it on a fresh database only.

4. Client

cd client
npm install

Optionally create client/.env. Without it, the client talks to http://localhost:5000/api:

VITE_API_URL=http://localhost:5000/api

npm run dev

Open http://localhost:5173 and log in with the admin account you created.

Environment Variables

Server (server/.env)

Variable	Required	Description
`MONGO_URI`	Yes	MongoDB connection string
`JWT_SECRET`	Yes	Secret used to sign login tokens
`GEMINI_API_KEY`	Yes	Key for alert, redistribution, risk explanation and chat text (fallback messages are used if it fails)
`PORT`	No	Defaults to `5000`
`PYTHON_SERVICE_URL`	No	Risk service URL, defaults to `http://localhost:8001`
`SIMULATE_CONSUMPTION`	No	Set to `true` to drain stock over time for demos

Client (client/.env)

Variable	Description
`VITE_API_URL`	API base URL, defaults to `http://localhost:5000/api`
`VITE_DEMO_EMAIL`	Shows a "Try the demo account" box on the login page
`VITE_DEMO_PASSWORD`	Password for the demo account (public, so use a read-only account)

Roles

Role	Can do
`phc_staff`	View the dashboard, alerts, redistribution and risk scores
`regional_admin`	Everything above, plus add and delete PHCs, stock, beds and attendance, and create users

Accounts are created by admins only. Use npm run create-user to create the first admin or to change a user's role or password.

API Overview

All routes except login and health need a Authorization: Bearer <token> header. ?lang=hi on /api/stock/alerts, /api/stock/redistribution, and /api/risk returns Hindi text; omit it (or use ?lang=en) for English.

Method	Route	Access	Purpose
GET	`/api/health`	Public	Health check
POST	`/api/auth/login`	Public	Log in and receive a token
POST	`/api/auth/register`	Admin	Create a user
GET	`/api/country`	Any user	List countries
GET	`/api/phc`	Any user	List PHCs
POST	`/api/phc`	Admin	Add a PHC (with bed counts)
PATCH	`/api/phc/:id/beds`	Admin	Update a PHC's bed totals/occupancy
DELETE	`/api/phc/:id`	Admin	Delete a PHC and its stock and attendance
GET	`/api/stock`	Any user	List stock records
POST	`/api/stock`	Admin	Add a stock record
PATCH	`/api/stock/:id`	Admin	Update a stock quantity
GET	`/api/stock/alerts`	Any user	Top stock-out alerts
GET	`/api/stock/redistribution`	Any user	Redistribution suggestions
GET	`/api/attendance`	Any user	List attendance records (includes patient footfall)
POST	`/api/attendance`	Admin	Add an attendance record
GET	`/api/risk`	Any user	Risk scores for the highest-risk PHCs (cached for 60 seconds)
POST	`/api/chat`	Any user	Ask the Gemini-powered assistant a question about current stock/alerts

Deployment

Frontend (Vercel): set VITE_API_URL, and optionally VITE_DEMO_EMAIL and VITE_DEMO_PASSWORD. Redeploy after changing them, because Vite reads them at build time.
API (Render): set the server variables above.
Risk service (Render): run uvicorn main:app --host 0.0.0.0 --port $PORT.
Database: MongoDB Atlas.

Made By

Parth Wakodikar — GitHub · LinkedIn

Navigation Menu

Repository navigation

README.md

README.md

MediSense AI — Smart Health & Supply Chain Resilience

Why Resilience

Demo Access (for judges)

The Problem

What It Does

Real / Realistic Data

How It Works

Google AI Studio & Gemini

Tech Stack

Project Structure

Getting Started

Prerequisites

1. Clone

2. Risk service (optional but recommended)

3. Server

4. Client

Environment Variables

Roles

API Overview

Deployment

Made By

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

MediSense AI — Smart Health & Supply Chain Resilience

Why Resilience

Demo Access (for judges)

The Problem

What It Does

Real / Realistic Data

How It Works

Google AI Studio & Gemini

Tech Stack

Project Structure

Getting Started

Prerequisites

1. Clone

2. Risk service (optional but recommended)

3. Server

4. Client

Environment Variables

Roles

API Overview

Deployment

Made By