Stefan Obradovic Email & Phone Number

New York, NY, US

College Park, Maryland, United States

• Accepted as Freshman after competitive application and interview process (5 of 80 applicants accepted)
• Developing portfolio backtesting framework and Quality Minus Junk asset pricing model in Python
• Applying portfolio optimization tools: Capital Asset Pricing Model and Fama-French Factor Model
• Conducting research into investing strategies and reading/analyzing relevant scientific publications
• Conducting research into time-series forecasting with Autoregressive Integrated Moving Average (ARIMA) and clustering with Dynamic Time Warping (DTW)
• Developing and integrating Machine Learning and Deep Learning strategies as part of the pipeline (Unsupervised Learning, RNN-LSTM, CNNs, XGBoost, etc.)

New York, New York, United States

New York, New York, United States

• Developed recommendation systems to improve song and podcast impressions on top of the home page (Shortcuts)
• Identified and presented differences in listening behavior that drive engagement across user devices and platforms
• Designed statistical experiments comparing deep neural network models for device-specific recommendations
• Created data pipeline for large-scale ML model development using Google Cloud, Bigtable, BigQuery, and Scala
• Deployed and evaluated online improvements of new production model with A/B test for 100 million users
• Helped migrate machine learning production code for home page from Tensorflow version 1 to Tensorflow version 2

College Park, Maryland, United States

• Working in the Capital One Machine Learning Research Stream under Research Advisor Dr. Raymond H. Tu.
• Scrum Master - Led research team of 4 students with scrum agile framework to design, implement, and apply a natural language processing machine learning model for automatic text generation.
• Performed data preprocessing, training, optimization, and evaluation using deep learning frameworks in Python (Keras, Tensorflow, NLTK, Gensim)
• Created a word vector-embedding model using a corpus extracted from thousands of news articles.
• Developed Recurrent Neural Network using Long Short-Term Memory to predict/generate a sequence of words in a user-given context.
• Presented research at FIRE Summit symposium poster session. November 15, 2019. University of Maryland, College Park

Mclean, Virginia, United States

• Analyzed credit card default risk for millions of customers and identified ways to help delinquent customers
• Developed software for customer segmentation to identify and understand the habits of risky credit users
• Created tools to mitigate credit loss during Covid-19 and understand changing credit card transaction behavior
• Interfaced with end-to-end big data pipeline for machine learning research using AWS, Databricks, and Snowflake
• Demonstrated results across enterprise, receiving high recognition from vice president of credit risk management

Philadelphia, PA

Ljubic, B., Roychoudhury, S., Cao, X., Pavlovski, M., Obradovic, S., Nair, R., Glass, L, Obradovic, Z. “Influence of Medical Domain Knowledge on Deep Learning for Alzheimer’s Disease Prediction,” Computer Methods and Programs in Biomedicine, vol. 197, Dec. 2020, 205765Developed program for data extraction using SQL queries, analyzed big electronic health.

Doctor Of Philosophy - Phd, Computer Science

Bachelor Of Science - Bs, Finance

Activities and Societies: Quantitative Finance Fellows, Smith Investment Fund, Business and Information Technology Society, International.

Bachelor Of Science - Bs, Computer Science - Machine Learning Specialization

Activities and Societies: Minor in Mathematics-Statistics, Computer Science Departmental Honors, Association of Computing Machinery.

High School Diploma

Activities and Societies: Pennsylvania Junior Academy of Science, Montgomery County Science Research Competition, Delaware Valley Science.