Alex T. Email & Phone Number

California, United States

Sunnyvale, CA, US

Premium AI Team.

Sunnyvale, CA, US

• Premium AI Team.
• Multi-task learning and graph neural networks for improving job recommendations.
• Generative AI for helping members improve their profile (generating headlines and summaries) and composing personalized messages to others (seeking work, general intro, etc).
• Deep learning for realtime personalized targeting of Premium upsells to drive Premium signups and revenue.

Sunnyvale, CA, US

• LTSC (LinkedIn Talent Solutions and Careers) AI Foundations Team.
• Worked on sequence modeling (CNN, LSTM, Transformers) for learning member activity embeddings to improve job recommendations.
• Leveraging multiple iterations of these embeddings in Job Recommendation models led to 5% increase in Predicted Confirmed Hires on LinkedIn.

Seattle, Washington, US

Course: CSE 312 (Probability & Statistics for Computer Scientists)Website: https://courses.cs.washington.edu/courses/cse312/22wiWinter 2022 quarter offering with 270 students, 20 teaching assistants, and partially remote and partially in-person.

Stanford, CA, US

CS109: Probability for Computer ScientistsCourse Website: https://web.stanford.edu/class/archive/cs/cs109/cs109.1218/Summer 2021 quarter offering with ~80 students and fully remote with lectures conducted over Zoom.

Seattle, Washington, US

• Course: CSE 312 (Probability & Statistics for Computer Scientists)Website: https://courses.cs.washington.edu/courses/cse312/20su/Slides, Videos, and Textbook: http://www.alextsun.com/prob_stat_cs.html
• Redesigned the course to include the fundamentals of statistics (e.g., estimators, confidence intervals, and hypothesis testing) and several applications to computer science in Python (e.g., Naive Bayes, bloom filters.
• Together with the teaching staff, developed new materials: a textbook, clear and concise slides, problem sets, autograders for nine new coding problems, concept checks, publicly available YouTube short lectures, and.
• Interviewed, hired, and managed a staff of 8 undergraduate TAs to ensure high-quality quiz sections, office hours, and materials.

• Course: Programming for Data Science Research
• Co-designed 6-week data science curriculum for high school students covering Python fundamentals, LaTeX typesetting, data science libraries (numpy, pandas, matplotlib, sklearn), and an introduction to machine learning.
• Interviewed, hired, and co-managed a staff of 10+ TAs to ensure high-quality discussion board posts and office hours, and efficient grading of assignments.
• Gave Zoom lectures daily to students from around the world, including "codealongs" where students coded with me during lecture.

Stanford, CA, US

• Course: CS 109 (Probability for Computer Scientists) Instructors: Chris Piech, Lisa Yan, David VarodayanHead TA during Winter 2020.
• Led the change to standardize coding assignments to use Python3, and developed autograders for assignments.
• Developed new section materials, homework problems, and exam problems.

Stanford, CA, US

• Course: CS 106A (Code in Place)Instructors: Chris Piech and Mehran Sahami
• Part of a teaching team for Code in Place, offered by Stanford during COVID-19 pandemic, with 10,000 global students and 900 volunteer teachers participating from around the world.
• Prepared and taught a weekly discussion section of 10-12 students to supplement professors’ lectures in a 5-week introductory online Python programming course based on material from the first half of Stanford’s.

Sunnyvale, CA, US

• Jobs Personalization Team under Careers AI.
• Worked on incorporating long-text semantic information such as member summary and job description into job recommendation pipeline.
• Experimented with fine-tuning BERT model using Tensorflow, but was computationally intractable due to the size of BERT and the dataset.
• Wrote pipeline to compute text embeddings offline with pre-trained BERT to use as features, resulting in a 1.3% AUROC lift on prediction of apply.
• Moved to CNN model with orders of magnitude fewer parameters than BERT, allowing for fine-tuning. Ability to fine-tune gave an edge over pre-trained BERT, resulting in a 2% AUROC lift on prediction of apply.

Sunnyvale, CA, US

• Jobs Marketplace Team under Careers AI.
• Worked on recommending daily budgets to job posters (advertisers).
• Implemented an end-to-end offline training and evaluation pipeline using an internal machine learning tool and Spark Scala.
• Investigated unexpected coefficients during the training phase using Spark notebooks and R to examine the data.
• Proposed a variation of Thompson sampling to generate more training data continuously.

Seattle, Washington, US

• Course: CSE 312 (Probability & Statistics for Computer Scientists) Instructors: Martin Tompa, Anna Karlin, Larry Ruzzo, Anup Rao, Adam Blank
• Awarded Bob Bandes Memorial Teaching Award, given to ~3 out of 400+ TA's each year.
• Prepared and gave lectures in professor’s absence to 100+ students.
• Prepared and led "advanced topics" sessions, on material not in the course, including: bloom filters, combinatorial theory, multivariate distributions, and more.
• Led quiz sections with ~30 students, wrote exam and homework questions, held office hours, etc.

Seattle, WA, US

• Washington Experimental Mathematics Laboratory (WXML).Working under Professor Jayadev Athreya on "Rotation Random Walks"
• Studying the behavior of random walks along a circle with a fixed irrational step size.Worked under Professor Sara Billey on "Graphs and Machine Learning".
• Goal is to create a database of graphs (nodes/edges) from the arXiv, in which people who currently are doing research that involves graphs can search by image and find papers (if any) that their graph(s) are contained.
• Have thousands of unlabelled images scraped from arXiv, and my work is to help classify the images which contain graphs so the features of the graph can be extracted and used in search. Uses convolutional neural.

Seattle, WA, US

• 1. Graphics and Imaging Lab.
• Did background reading and learning on computer vision, deep learning, and convolutional neural networks (CNNs); wrote scripts for part of an image pipeline to feed into the neural networks. Wrote Gaussian Naive Bayes.
• Overall Goal: Will use CNNs for classifying emotion separately for human images and for animated character images. Then, will find some mapping from the human feature vector to the animated feature vector (second to.
• Currently reading papers about adversarial machine learning and learning theory. After exploring several papers, we plan to formulate some new problem statements and answer them.

• Product Navigation Team under Core App Pillar.
• Implemented a pipeline to regularly run PCA and K-means clustering on a large dataset, using Hive, Presto, Python, and other internal tools.
• Performed several other tasks, including fitting linear/logistic regression models, to find insights and give business recommendations.

Mountain View, CA, US

• Flexible Creatives Team under Search Ads.
• Designed and implemented a multi-stage data pipeline starting from collecting data from ads serving logs to computing a score used for evaluating creatives within an ad group.
• Based on these computed scores, possibly recommend to advertisers which creative(s) to remove from the ad group in order to boost performance of the ad group and increase revenue for both advertisers and Google.
• Implemented the MapReduce job using SQL and the Flume framework in C++. Other technologies used include Dremel, SSTable, and ColumnIO.

Master Of Science - Ms, Computer Science

Bachelor Of Science - Bs, Computer Science, Statistics, And Mathematics

High School Diploma