Sivar Azadi

Machine learning engineer with Physics BS from the University of California, Santa Barbara (UCSB) exploring data science and machine learning roles.

Summary

Sivar Azadi is a graduate from the University of California, Santa Barbara’s (UCSB) College of Letters and Science, where he received a Bachelor of Science in Physics with departmental honors. In addition to his background in physics, Sivar has also honed his skills in machine learning and data science through his experience at UNDO Carbon, the LUX-ZEPLIN (LZ) dark matter experiment, and various personal projects. Sivar is currently seeking new opportunities in the field of data science and machine learning.

Experience

Machine Learning Engineer Jan 2023 - Present

UNDO Carbon

Conducted a literature review and developed predictive models for flood detection and river-to-ocean chemistry changes. Utilized PyTorch and hydrology ML packages to implement LSTM and GRU models to improve forecasting accuracy and efficiency. Additionally, led the development of a white paper on the potential of ML for global implementation of enhanced rock weathering.

Particle Physics Research Analyst Sep 2020 – Sep 2022

Lippincott Lab - UCSB

Implemented deep neural network models using TensorFlow to classify Kr83m and Deuterium-Deuterium particle interactions with 98% accuracy. Created an automated data quality monitoring tool for the skin detector that saved hundreds of hours of independent analysis from 300+ researcher members. This allowed for improved distinguishability of detector sensitivities and facilitated the dark matter search.

Projects:

Identifying and Outlining Breast Cancer Tumors

Developed a deep U-Net model for image segmentation to detect cancer in ultrasound images, resulting in a validation accuracy of ∼ 95%, Mean IoU of 0.735, and F1 score of 0.793 on the test set. Visualized true masks and model masks to compare the predicted location of cancer with the ground truth annotation, contributing to cancer recognition

AI Art Curator: Binary Classification

Built a binary classification model that can classify artworks in different art movements using image augmentation techniques and pre-trained ResNet50V2 model. Scraped the dataset using 'WikiArt Retriever' by Lucas David and uploaded to Kaggle for public use

Spotify Song Grouping with K-Means Clustering

Implemented data preprocessing techniques, including cleaning and normalizing values, on a dataset of Spotify songs to prepare for clustering using K-Means algorithm. Evaluated the quality of the resulting clusters using silhouette score evaluation metrics and visualized the clusters through 2D projections, coloring data points according to cluster assignment

Language Detection with Deep Learning

Built deep learning module for text-to-language detection, with 18 languages and 97.9% accuracy, using Language Detection and 4000+ Kurdish words dataset from Kaggle

Text From Image Recognition with Convolutional Recurrent Neural Network (CRNN)

Developed model to recognize text from images of common English words employeeing keras tuner for hyperparameter tuning. This model was trained on the combination of the Synthetic Word dataset on kaggle and a character level image dataset.

Boston Crime Mapping and Analysis

Created several mappings of crime geospatial data in Boston. Performed analysis of interesting features of data such as crime over time using time-series and most commited crimes.

Ongoing Projects:

Dynamic Object Tracking

Deep learning-based computer vision project for real-time object tracking in videos

Data-Driven Justice: Supreme Court Predictions

Developing predictive model using NLP techniques to determine the outcome of Supreme Court cases from a dataset of 3304 cases from 1955 to 2021

Cost of Living Insights: A Global Analysis of Urban Economies

Analysis of global urban economies using a dataset of over 4500 cities to provide valuable insights on cost of living trends, utilizing data visualization and statistical techniques

Certifications:

Coursera Deep Learning Specialization

Relevant skills: Artifical Neural Networks, Hyperparameter Tuning, Recurrent Neural Networks (RNNs), Convultional Neural Networks (CNNs), LSTMs

DataCamp Associate Data Scientist

Relevant skills: Statistical Experimentation, Data Management in Python, Exploratory Analysis, Communication and Visualization

AWS Associate Solutions Architect Course

Relevant skills: Cloud Design and Deployment, EC2, S3, AWS Automation Tools, VPC, RDS, Disaster Recovery and Availability, Route 53

Bachelor Thesis

Offline Data Quality Monitoring for Skin Detector of LUX-ZEPLIN (LZ) WIMP Detector

Presents the work done in maintaining the LZ health module, tuning it as commissioning data is being taken to ensure data quality, and the various independent analyses accompanied in developing and understanding the Skin as an effective veto system.

About Sivar

Sivar is an aspiring data scientist/machine learning engineer who loves delving into the possibilities of AI and machine learning. He finds the idea of creating solutions and models using machine learning extremely captivating, and this passion has led him to pursue a career in the field. Sivar's portfolio demonstrates his skills in developing efficient and accurate models, working with large datasets, and his drive to find new ways to apply machine learning to real-world problems.

Contact

sivarmazadi@gmail.com