Yingjie (Jane) Qu

Experience

• 

  微软

  May 2021 - Aug 2021

  Support Engineer Intern

  • Conducted user funnel analysis in SQL to gauge user participation levels at each stage of the Intune UX funnel, pinpointing 3 main challenges in user experience.• Implemented multiple C# projects to improve Intune user login authentication, user notification, and device management functions via Net Framework from data insights, which enhanced system reliability by 21%.

• 

  Civilience

  May 2022 - Aug 2022

  Data Science Intern

  • Web scraped 50K+ Covid-19 tweets via Snscrape Python library and encoded with SentenceTransformers, extracting highly correlated topics for future subsequent clustering.• Implemented Topic Modelling and K-Means to categorize and label 50K+ tweets into 200 distinct topic groups to better understand user engagement and interests in various topics of Covid-19.• Constructed a Cloud-based ETL pipeline with AWS EC2, Lambda and EventBridge, which automatically curated the Top 200 popular daily topics among 50K+ tweets, achieving a 10% increase in operational efficiency.• Visualized comment counts as key metrics in 2 Tableau dashboards with 3 sheets to replace manual slide presentations, increasing report efficiency by 20% and achieving 10% boost in click-through rates from 65% to 75%. Show less

• 

  HCL America, Inc.

  Jan 2023 - Apr 2023

  Talent Analytics Intern

  • Implemented a keyword search algorithm in Python as a baseline to assign 20K+ employee comments with 5 topic labels, getting an average hamming loss of 0.3 and saving an estimated 40 hours of work for manual labeling.• Preprocessed the 20K+ comments via Python by removing stop words, transforming text data into numerical matrices, splitting data into training and test sets, and applying oversampling method to address unbalanced topic labels, which reduced the MSE of preceding models by 26% from 0.27 to 0.2 and eliminated overfitting. • Developed machine learning and deep learning models for automatic topic labeling on 20K+ employee feedback, achieving an average hamming loss of 0.1 with a fine-tuned RoberTa model using the SimpleTransformers library and 0.12 with a multilabel-classification model using the Skmultilearn library, which resulted in a 60% to 67% decrease in average hamming loss compared to the baseline model; Identified the multilabel-classification model as the optimal solution for adding labels based on comparably low hamming loss and high operation efficiency. Show less

• 

  Analog Devices

  May 2023 - now

  • Manufacturing Engineer

    Jan 2024 - now

  • Foundry Engineering Data Analysis Intern

    May 2023 - Dec 2023
• 

  Unilever

  Sept 2023 - Dec 2023

  Student Data Scientist