Aegis AI

Product Dashboard
Confusion matrix + Histogram of continuous thermal comfort scores

Problem and Users

In the United States, heat stress causes an enormous human and financial cost each year, resulting in the deaths of hundreds of workers and more than $1 billion in medical bills. Our investigation shows a serious weakness in the companies present standards as most current safety practices are reactive and compliance-driven, missing real-time insight into worker health. Workers are left susceptible by this compliance-focused strategy's failure to include real-time health monitoring. Beyond safety, there are serious operational inefficiencies as a result. Unmanaged heat stress, for example, is directly responsible for productivity losses in the construction industry that range from 29.0% to as high as 41.3%. This underscores the urgent and evident need for a preventative solution.

Data Sources

Heat stress impacts workers and the bottom line | Harvard T.H. Chan School of Public Health WHO, WMO issue new report and guidance to protect workers from increasing heat stress Occupational Heat Stress and Kidney Health: From Farms to Factories - PMC 5TH ABC Challenge: Forecasting Thermal Comfort Sensations for Heatstroke Prevention: Leveraging Physiological Data for Better Outcomes | IEEE DataPort

Model/Approach

To build our solution, we combined a public IEEE dataset on Thermal Comfort (age, gender, body temperature, HRV) with environmental sources, including real-time weather APIs that provide humidity, heat index, and air quality data. We grounded our thresholds in OSHA safety guidelines and developed predictive models learning from worker behaviors over time using a smart wearable that generates individualized “Heat Risk Scores.” These scores drive automated alerts: a yellow alert signals caution, while a red alert instructs the worker to stop immediately and hydrate. The data aggregates into a dashboard that supports compliance reporting, scheduling, and liability reduction so companies are better able to analyze their worker health, preventing heat exhaustion and stroke before they occur.

Deployment Plan

The backend is hosted on an AWS EC2 instance, running a Flask web server with Python 3. The frontend is deployed on Vercel and built with Next.js, React, Tailwind CSS, and TypeScript. The machine learning model, based on Random Forest (XGBoost), is designed to deliver high accuracy in predicting worker safety risks.

Impact Metrics

Productivity Boost (Downtime Hours Prevented), Less Insurance Payments (Workers' Compensation Claims Avoided), More Workers Lives Saved (Predicted Fatalities Averted).

Data Governance

We ensure compliance by de-identifying all data before use, enforcing strict PHI handling through encryption, access controls, and retention policies, and aligning with partner licenses and ToS. The ML model is documented with a transparent model card, including intended use, limitations, and fairness audit results. Regular bias checks and governance reviews safeguard equity, privacy, and legal compliance across all systems.