Tarik El Lel

The project required the development of data models and ETLs to enable analyses of customer lifetime metrics over time, including cohorts, retention, and repurchase metrics. The project included designing models and implementing them in the data build tool (dbt) and BigQuery and scheduling them using Apache Airflow. Data visualization modeling was then done using LookML on Looker. The processes allowed measuring customer lifetime value over time and understanding how the retention metrics changed across different markets.

Built a notification service that would pull daily customer data, parse any relevant behavior, and generate a ticket for customer service to action. The project included the addition of A/B testing to the service that measured which actions would work best. ETL was created to record all notifications or tickets and to measure experiment success. The stack used on the project was Google Compute Engine, Docker, Airflow, BigQuery, and Salesforce.

This python-based service would recommend user sizes on product listing pages based on their previous shopping history to minimize size-based returns. I worked with the data science team to build the ETL to prepare data for ingestion by the deep learning algorithm for the daily training of the model. I also designed one of the prediction algorithms, which used Word2vec for feature generation and random forest for prediction.

The project was a business-to-business ETL pipeline from a source database and weblog. It ran containers for the source transactional database, destination analytical database, Airflow scheduler, and a lightweight database client, resulting in a source Postgres database model for the customer. The project also included the establishment of a destination database, pipeline, and model to meet the customers' analytics requirements, including an ELT from the source Postgres database to the destination Postgres database.

As a part of the final project during my master's degree, I designed an ensemble deep learning model that used three convolutional neural networks and a classifier ensemble layer to detect whether a chest X-ray belonged to a patient with COVID-19. The model achieved 95.66% accuracy.