Donya Hamzeian

(ehsAI was acquired by Intelex) - End-to-end ML pipelines development and maintenance - Implement proof of concepts for multiple projects - Provide AI solutions to meet customers needs - Fine tune Large Language Models (LLMs)

(Startup company later acquired by Intelex, team size: 4-5 ML engineers) - using AI/ML to deconstruct EHS documents - using AI/ML to extract structured information from unstructured data - maintaining ML models in production

• Assisted undergrad and grad students in online or in-person office hours and via piazza • Proving cross-cultural communication skills, given the diverse environment at UW • Prepared solutions to the assignments

• Scraped and preprocessed data from "trip advisor website" • Performed sentiment analysis using multinomial naive Bayes method in python, resulting in approximately 90% prediction accuracy • Discovered influential amino acids in the Vitiligo disease using statistical methods like logistic regression, multiple hypothesis testing, and random forest in R. Also, generated data visualization for exploratory and explanatory data analysis using ggplot2

• Extracted information from a semi-structured ancient text resource using Scipy, Numpy, Gensim, NLTK, Spacy, Hazm, and so on in python. • Created the first database in the Iranian traditional medicine using postgresql and SqlAlchemy • Successfully collaborated with non-statisticians from the medical fields • Co-authored a journal article about the database: https://www.hindawi.com/journals/ecam/2020/3690781/

• Facilitated 3 workshops in data visualization and text mining in R for Phd students • Prepared workshop materials by consulting multiple resources

The company is an investment consultancy company. As a member of the programming team, I contributed to developing a back-test engine for trading algorithms. I learned C# and MSSQL by this project.

Here is the abstract of my thesis: "Exploratory literature review of the COVID 19 papers has become a time-sensitive and ex- hausting challenge during the pandemic. The novel topic modeling pipeline that we present in this thesis helps biomedical researchers in having an overview of the topics existing in the papers. The preprocessing framework that we present captures the UMLS entities by using MedLinker which handles Word Sense Disambiguation (WSD) through a pre-trained BERT model. The model that we used is a Variational Autoencoder implementing ProdLDA, which is an extension to the LDA model. Applying our framework to the CORD-19 dataset, we were able to achieve a topic coherence value of ∼ 0.7 and high topic diversity. Our pipeline can be used in other biomedical texts as well. Also, it can make an inference about the topics of a document with unseen vocabularies via MedLinker linking the new word to the UMLS entities."

I have studied computer science as my major and economics as my minor. Studying in this university consolidated my analytical ability and mathematical skills. I graduated as the 1st rank among all the BSc students of the CS department.