Sepehr Ardebilianfard

Software Engineer

Capital One•  June 2020 - August 2020

• Designed predictive models with Keras by utilizing GRU and attention architectures. Given graph embeddings of account transactions model, predicted new monthly embeddings to be used for spending patterns and anomaly detection. • Given the large size of embedding data, a data pipeline was implemented with parallel processing on EC2 to decrease training and testing of time by 40%.

Software Engineer

Retech Labs•  November 2018 - November 2019

• Created a visual-based barcode generator that increased our image recognition model's baseline precision by 73% to identify grocery products that lacked differentiable visual features with previous precision of 20%. • Implemented Object Detection and feature extractor models across client retailers, helped achieved 98% accuracy on client data by developing various accuracy tools built with TensorFlow, OpenCV, and several other ML libraries. Used Adaptive Threshold algorithm to find false positives in model’s prediction to improve accuracy by 5% with 94% baseline accuracy and delivered a contract by passing 98% model accuracy for the client. • Managed a potential client from pilot phase through production, provided technical support for clients during the initial phase, followed staff training on data collection and validation.

University of California, San Diego (UCSD)

Computer Science & Engineering, BS•  September 2018 - Present

Studying Computer Engineering and experience in Machine Learning and SWE