Bilal Mahmood

Built a deep-learning-based image classification model using transfer learning to measure whether X-ray images belong to COVID-19 patients. The project was part of my master's course. The model showed high accuracy on the validation set.

I created a prototype for an insurance recommender system using machine learning classifiers that accurately and transparently ranked customers to be targeted with property, disease, accident, and legal insurance.

RESPONSIBILITIES
• Trained several supervised machine learning classifiers using the Tree-based Pipeline Optimization framework and selected the best explainable model using the cross-validation technique, where the AUC score was the optimized metric.

• Transferred machine learning methodology to the Sparkasse team by guiding them on the essential steps required to build ML models, from data preprocessing to feature engineering, feature selection, model building, and model evaluation.

• Produced a detailed research report at the end of the project to communicate the critical findings at each level of the model-building process.

• Deployed the model building and inference pipelines in the Konstanz Information Miner framework for transparency and a better understanding of the Sparkasse team.

I translated a complex credit score rating algorithm (for a client for a client, more information at the link below) that measured the loan worthiness of different customers into efficient Python code and deployed it as an AWS Lambda function. It was exposed with a REST API endpoint to be consumed by the front-end team to display the ratings.

I also orchestrated the rating calculation using AWS Step Functions. When customers upload their financial documents into the S3 bucket, that event triggers the parsing of the files. It then prepares the data consumed by the rating calculation endpoint, which stores the results in the DynamoDB database.

URL: https://fdg.mcc.it/rating