HyperWorks AutoML

This project was developed for automating data analysis pipelines in material science labs. We primarily targeted battery manufacturers, solar cell manufacturers, and research labs in large universities. The web-app was a no-code tool for auomatic feature engineering, model training, and ensemble learning to give the most generalizable model that satisfied task requirements. A custom backend was provided for battery cycle-life prediction where domain experts could choose a subset of important features for testing battery cycle-life hypotheses.

Challenge:

Developing a Full Stack Web App from scratch with no prior background.

Motivation:

Material science researchers did not necessarily have exposure to data analytics and ML methodologies. Moreover, existing autoML tools back then, needed setting up of programming environments and required familiarity of a wide range of ML tools. This app provided a no-code solution to researchers to quickly test chemical compositions and physical parameters for their processes with a user friendly UI.

Tech Stack:

• Backend server: Django
• Server Load Balancing + Reverse Proxy: Nginx, Docker
• Databases: MongoDB
• App Development: Angular 8, Highcharts

I gained a lot of insights in software development by learning and implementing frontend, backend, and CI-CD pipelines from scratch. My expertise in ML eased the actual machine learning aspects of the project, however, syncing them with a website, plotting interactive graphs, and ensuring all the async elements work with each other was a very enriching experience.