iSchool Capstone

2020

Project Logo

ML Enabled Predictive Quality

Talking Rain(TR) is a leading beverage company manufacturing sparkling water and other beverages. TR relies on manual quality assurance processes to identify non-optimal beverages before release into the market. Non-optimal products yield complaints from consumers and generate negative publicity for the company. The goal of the project was to leverage ML to identify hidden patterns and detect non-optimal lots to prevent their release into the market. Our ML model was able to identify 100% non-optimal lots at the cost of retesting few good lots. This data-driven approach has enabled TR to identify quality issues at co-packers early, ensuring proactive remediation.
Project Logo

ML Enabled text summarization

Creating an abstract from claims in patent application is a time-consuming and tedious process, which requires manual effort. Often times, the patent attorney has to go through pages of claims to form 150 words abstract. It’s a monotonous process that has good scope for automation. The time saving achieved from automation can be utilized in more challenging issues. We explored three different models for our project: PageRank, LongShortTerm Memory and Text to text transform transformer(T5). The T5 model has shown the most accurate result so far and helped us in generating meaningful and grammatically correct abstract without any human intervention.
Project Logo

Moonshot

The Pacific Science Center collaborated with Magic Leap to provide unique and immersive experience to visitors. Project Moonshot is a game application that utilizes Magic Leap One’s spatial computing ability to bring the entire solar system within ‘reachable’ distance to the players. Player’s task is to place planets in the correct orbits respectively. To ensure the accuracy and relevance of the educational value, the game is designed based on Common Core states standards for grade 6-8 students. While the game application is immersive, it is also interactive which requires multiple players to collaborate and exchange information in order to win.
Project Logo

Patent Evaluator

Patent applications are currently an expensive process, requiring many work hours and high economic costs for creating, editing, and submitting applications to the United States Patent and Trademark Office. This project utilized machine learning methods to analyze method claim text and ultimately determine whether a method claims application would be rejected. This project aims to help stakeholders leverage the results from these models to streamline their patent application process and to implement the models in the future for better determining which words may contribute to their application being rejected.
Project Logo

Project Detox

Among gamers playing the top 15 games, roughly 70% have experienced online abuse. Project Detox is empowering the Gaming Safety team at Microsoft so that they can provide a safe environment to their customers worldwide. We are implementing an automated testing framework for their toxicity classifiers. After measuring model performances against each other and on different kinds of data, we have generated beautiful and intuitive reports which would enable the stakeholders to make data-driven decisions. All of this has been packaged into pipelines which would automate the process and eliminate manual work.
Project Logo

PUSHSPRING Marketplace Search

Our marketplace search recommender system is a web app that users can use to explore the apps related to a persona or a given mobile app based on PushSpring's local user data rather than average result such as the ones Google gives. Our system was implemented with both CNN and NLP to maximize diversity and relevancy in recommendations. Combining benefits from both content-based and user-based recommendation models, our model produces valuable insights that can't be found elsewhere.
Project Logo

Reinventing Students’ Career Navigation Experience

Landing a good job and having a well-planned career path is necessary for every college student or graduate. While during this process, connecting with seniors or networking plays an important role to figure out the career path. But recently, some discovered facts show that many students don’t realize the value of networking and don’t know how to manage the networking process. This project aims to create a holistic user evaluated digital experience that could improve students’ experience and journey of choosing from their career options and appropriating steps they should take to reach their goals.
Project Logo

Scheduling Software Pilot for Seattle Public Library

Employee scheduling is a crucial task for most organizations regardless of their size. Currently, the Public Services team at SPL manually prepares the schedules of their employees using Microsoft Excel and Sharepoint. This manual process takes a lot of time and effort to create as well as manage schedules. We conducted research to build a flexible solution providing user-friendly interface and easy customizations. This solution would enable the scheduling team to be agile in creating and changing schedules as per the library branches’ requirements, thus enabling resource optimization and quick decision making.
Project Logo

Setting Up a Checkpoint for Research on Journal Data Policies: A Systematic Review

To ensure the transparency and reproducibility of scientific discoveries, more and more journals have adopted data policies to encourage data sharing. Many previous studies have investigated the prevalence of journal data policies (JDPs) in certain disciplines. However, it is still unclear 1) how the prevalence of JDPs has changed over time and varied by discipline; and 2) when trying to interpret JDPs, what aspects researchers are most concerned about. This capstone project reviewed 27 relevant publications and built a toolkit that contains a dataset and a meta-analysis report so users can easily understand the development of JDPs.
Project Logo

The modern product manager’s toolkit! Sentiment Analysis and Topic Modeling

Smartsheet is an enterprise platform that helps automate processes across a broad array of use cases, which strives to improve its offerings based on consumer feedback. The task of understanding and leveraging unstructured comments and emails from customers is challenging, which is why we propose a combination of topic modeling and sentiment analysis techniques to gather actionable insights, aggregated and presented as a dashboard to stakeholders thus enabling them to take data-driven decisions to improve user experience. The solution focuses on identifying negative feedback from users, in order to highlight potential areas of improvement, thus informing the future product roadmap.