Case study 03 Full-stack · Machine learning

Hospital Operations Sync Platform

One live view of a hospital's beds, queues and stock, with models that forecast what runs out next.

Role: Full-stack engineering and machine learning
Timeline: Jan 2026 · hackathon build
Stack: Django REST Framework, React, MySQL, scikit-learn, JWT, Razorpay
Recognition: Hackathon project

Admin dashboard showing total and available beds, OPD patients, active admissions and low-stock alerts

ML models in the product: 6
database tables: 33
role-based dashboards: 4
feature modules: 11

Overview

A hospital operations platform that gives doctors, nurses, administrators and receptionists their own live dashboards (beds, OPD queues, admissions, inventory and billing) backed by machine-learning models that predict wait times, stockouts and financial risk.

The problem

Hospital staff juggle beds, queues, stock and billing across disconnected tools. Shortages and long waits are usually discovered after they have already happened, and neighbouring hospitals have no quick way to share capacity.

The approach

Put every operational signal into one platform with a dashboard for each role, then add models that look ahead: how long the next patient will wait, which medicines will run out, and where money is being lost.

AArchitecture

How it fits together

01Clients

React dashboards Doctor, nurse, admin and receptionist
City capacity view Anonymised bed and ICU availability

02API

Django REST Framework 11 apps behind JWT authentication
Payments Razorpay orders, verification and webhooks

03Intelligence

6 scikit-learn models Wait time, stockout, profit and loss
Weather demand engine Weather + air quality → medicine demand

04Data & services

MySQL 33-table relational schema
OpenWeatherMap Current weather and AQI, cached for 1 h

BEngineering decisions

The calls that mattered

01
Predictions where they change a decision
Instead of one showcase model, each prediction sits where staff act on it: wait-time estimates in the OPD queue, stockout risk and days-to-stockout in inventory, and profit and loss-area forecasts on the billing dashboard. The wait-time estimate falls back to a rule-based calculation if the model fails, so the queue never goes blank.
- OPD wait-time regression
- Stockout classification + days-to-stockout regression
- Profit and loss-area prediction
- Rule-based fallback for wait times
02
Weather-aware medicine demand
Live weather and air-quality readings from OpenWeatherMap feed a rule-based engine that maps conditions to likely disease spikes and recommends stock increases, such as extra antibiotics and inhalers when respiratory infections are likely. Responses are cached for an hour, with mock data when no API key is configured.
- Current weather + AQI
- Condition-to-disease mapping
- Recommended stock adjustments
- 1-hour response cache
03
One platform, four roles, real payments
Django REST Framework serves 11 feature apps behind JWT authentication over a 33-table MySQL schema. Each role gets its own dashboard, a public city view shares anonymised bed and ICU availability across hospitals, and billing runs through Razorpay with HMAC-SHA256 signature verification on every payment.
- 11 Django feature apps
- 33-table MySQL schema
- JWT authentication and role-based dashboards
- HMAC-SHA256 verified payments

CIn the product

Screens

Inventory view with ML stock predictions flagging items that need reordering soon, with days left and risk scores — ML stock predictions with days left and risk scores

ML wait-time prediction of 78 minutes for the OPD queue, with confidence, queue status and factors considered — OPD wait-time prediction with the factors behind it

Recommended stock increases for respiratory-infection medicines driven by weather conditions — Weather-aware medicine demand recommendations

Receptionist dashboard with billing totals, predicted profit and a loss-area risk indicator — Billing dashboard with profit and loss prediction

Bed management grid showing available and under-maintenance beds by department — Bed management across departments

Appointments management screen with pending approvals and today's appointments — Appointment approvals

Outcome

Built end to end as a hackathon project: a working platform covering OPD queues, beds, admissions, inventory, inter-hospital sharing and billing. The full source is on GitHub.