Marcus Pop

Contributed to a digital wallet application connected to the national banking infrastructure through the Automated Clearing House and instant payments, tokenization, and central banking systems.

My responsibilities involved:
• Implementing various endpoints.
• Building the data model layer.
• Developing an administration system for multiple entities using the four-eyes principle.
• Creating a one-time password system for logging in.
• Assisting with designing and creating the KYC back-end flow for the mobile application.
• Connecting various systems to our back end.

I implemented a dialogue manager module inside a currently in-development minimalistic voice assistant application. It enables users to communicate with it in Romanian and issue simple voice commands. The dialogue manager is a component inside the project for interpreting given commands and their parameters, deciding whether they are correctly specified or not; retrieving information from APIs, such as the weather or calendar; and issuing a meaningful response based on a given input.

I implemented a web application for the Microsoft Imagine Cup 2019 using Node.js and Azure. The application intends to give performance feedback to educators teaching children with disabilities. This feedback is based on emotion detection and analysis of incoming heart rate data and allows educators to create a lesson plan based on acquired and interpreted data.

Implemented various models for semantic segmentation using UNet and Fast-SCNN and trained them on the Cityscapes Dataset to improve instance-based segmentation. Used other deep learning models, such as Mask R-CNN and Deep SORT, to construct an end-to-end computer vision pipeline to segment and track people from video files.

Preprocessed a raw dataset about early COVID-19 patient data, visualized it, and applied different methodologies to deal with unbalanced data to classify the disease outcome using logistic regression.

A data science project aimed to classify fraudulent transactions given an unbalanced publicly available dataset. Had to engineer a lot of new features and clean and deal with missing data. I also had to construct plots and measure statistics to draw insight from the dataset. Finally, I have built several classification models to find the best one for the task (neural networks, decision trees, and logistic regression).

I designed and implemented the authentication between two microservices and RESTful APIs using Keycloak. The application was deployed using Docker Compose and a common network. Requests were made using OpenFeign for easier declarative API calls, and the security was handled using Spring Security and Spring OAuth 2.

Integration of OpenAI's ChatGPT API in existing chats for document-based tasks using available SDKs, Langchain, and AWS. The final purpose is to create a plug-and-play solution that can easily be integrated into various existing popular chat systems (Slack, Microsoft Team, Mattermost), available through a monthly subscription.