I joined AppFolio's Dynasty team to build AI Leasing Assistant — Lisa. She helps prospective tenants schedule showings via SMS or email.

Here are some of stuff I implemented during my summer internship at AppFolio. I've generalized the ideas so I won't violate NDA.

• Linkification: Given a list of keywords, linkify the keyword in a message. For example, let's say the keywords are { "apple": 🍎, "grape": 🍇 } and the message is I am eating an apple. Then the output should be I am eating an 🍎. To achive this, I

  1. split the message by keywords using regex (e.g. I am eating an apple → ["I am eating an ", "apple"])
  2. Replace the keywords with emoji (e.g. ["I am eating an ", "apple"] → ["I am eating an ", "🍎"])
  3. Join the array (e.g. ["I am eating an ", "🍎"] → I am eating an 🍎)

• URL Encoding/Decoding: Every time we create a zoom meeting room, it's an unique URL (i.e. something like zoom.com/kfjaskrdakjahfkjhad). But we want to provide the same user the same URL every time. For example, user with id foo may just visit zoom.com/foo. My job is to check the database and redirect the user to the correct URL under the hood. The idea is to have the user to bookmark their link so they don't have to confuse themselves from using the wrong link.

• Slack Integration: When an user logs in to the website, send the message to a Slack channel to notify everyone who just logged in.

• GraphQL Subscription: Our web application was providing real-time data to the users with polling. This has been putting a lot of pressure to our server. Integrated GraphQL subscription to our application so that the server will send the updated data to the client only when needed, effectively reduce the load of the server. This was the most challenging project I did during the internship as it involved a deep understanding of computer networking as well as AWS infrastructure.

• Email Classifier: Given an email, we want to classify whether the message is about leasing or not. We already had labeled data in the database. I extracted features such as number of token and number of leasing keywords from the message. Then I trained a logistic regression model with scikit-learn.

I learned a ton from my engineering manager John Rothfels during this internship. One valuable skill/hack I learned is that not everything is documented. Sometimes we need to dig in to the source code of a library to solve a problem.