• Natural language processing and collaborative filtering techniques can be applied to many different domains. Through my projects I learned about the ins and outs of deploying a machine learning model at scale and the importance of tailoring the model to the problem at hand.

Saltiest Hacker News Comments

• A week long project that involved creating a NLP model to find the “saltiest” comments on the Hacker News website using NLP.
  • My contribution: applying Natural Language Processing techniques to text and populating of database of predictions to serve to the front end.
  • Explanatory post: Sentiment Anaylsis with the flair NLP library
  • Repo for the project

Spotify Song Recommender

• A week long project that involved deploying the DS API for an app that would recommend songs based on Spotify data.
  • My contribution: deploying a predictive model with Flask and hosting it on Heroku
  • Explanatory post: Tips for deploying a simple model with Flask on Heroku
  • Repo for the project

Groa - Movie Recommendations

• A two month long project that involved deploying an existing model into a new backend infrastructure.
  • My contribution: migrating an existing predictive model for movie recommendations to a automatically scaling AWS insfrastructure with the Cortex library
  • Explanatory post: Deploying a Recommendation Engine with Cortex
  • Main Repo for the Groa team
  • My repo with the code used for Cortex deployment

• My own extension of the project where I train a BERT/Transformer model on the Groa movie review data to make sequential predictions.
  • My contribution: applying a Pytorch implementation of a transformer based model with bidirectional encodings to user watch histories to get movie recommendations.
  • Explanatory post: BERT for Movie Recommendations