Theodoros Goulas

I developed a deep learning model that performs anomaly detection on tabular and unstructured data or text extracted from financial messages under ISO 20022. The model uses variational autoencoder architecture to detect anomalies and the SHAP model to determine the key factors that explain the model's decisions. The application is built and integrated with the AWS platform (Amazon SageMaker).

I built a BI web application for customers of Cardlink, Greece's largest card acceptance network. I was mainly assigned back-end development tasks, incorporating Spring Framework microservices for Java 8. I also handled some front-end tasks in React and Node.js, mostly regarding the integration with external services, enabling the payment process using the PSD2 protocol.

A multitenant cloud application built on the AWS cloud platform can parse financial messages, such as SWIFT MT, ISO 20022, and SEPA, etc., and provide AI and ML metadata analysis of extracted data. This application consists of two distinct components: admin application and tenant application. The admin app is hosted in the management AWS account and handles tenants' onboarding, retention, monitoring, and billing processes. The tenant application is hosted in separate AWS accounts under the same organization. It receives bulk and real-time financial messages and provides ML insights and BI tools for statistics, KPIs, visualizations, and reports using Amazon ML services, SageMaker for forecast and anomaly detection, and external frameworks like GraphSAGE and a graph neural network for fraud detection.

A large-scale custom web application that handles user management and billing processes of Cardlink, Greece's largest card acceptance network. I implemented a set of RESTful APIs based on the Spring Boot framework and batch processes handling Cardlink's billing processes.