eBay HelpBot

Work Project. Live now in the UK for hundreds of thousands of users, the HelpBot is a product backed by a machine learning-based natural language understanding engine to take care of issues on the member’s behalf. For instance, users can submit returns, get refunds, or have disputes with sellers adjudicated without even having to leave the page. Link below goes to one example help topic--HelpBot is available on many more topics where chat is available. Click "Chat with us" at the bottom of the page and make sure you're signed into eBay.

Work Project. Live now for a portion of members in the UK. We launched a routing system utilizing the HelpBot's natural language understanding capabilities to reduce contact transfer rate in eBay's contact centers (transfers are a bad experience for the user, and expensive for eBay). This routing application identifies the user's issue from a brief description and connects them with the correct staffgroup.


Work Project. Launched a new machine learning classifier-based search engine for the eBay Help platform in Australia, which is seeing a large improvement over the prior Oracle-based keyword search engine in terms of click-through rate, average click position, and search comprehension.


Mission Curated

Personal Project. I co-created this web app with Fiona Foster as a love-letter to our neighborhood in San Francisco.

Bay Area Worker's Guide to Backpacking

Personal Project. Co-editor of this Medium publication aimed at similar working folks in the Silicon Valley/San Francisco area who love backpacking like we do. Our goal here is to boil down excursion summaries and insights to the bare essentials.

eBay Document Upload Tool

Work Project. Led the build for the redesigned document uploader tool for eBay customer service requests. Rolled out to all eBay sites.


Personal Project. Based on the tutorial by Siraj Raval with quite a few modifications. This convolutional neural network model is trained on the MNIST dataset using the Keras API and TensorFlow as a backend. The application itself is deployed on Heroku, using Flask and Gunicorn to serve.