ChillPill

ChillPill repository on GitHub

As an Insight fellow, I built ChillPill, a web app that recommends anti-depressants based on patients’ symptoms and the most common side effects from taking pills. Rather than recommending anti-depressants for therapeutic purpose, I aimed for this website to be used so that patients can further consult with their doctors. In summary, I

• Extracted side effects from patient survey data using topic modeling, and infered their prevalence from the corpus of patient experience
• Engineered features from 17,000 subreddit comments collected through Pushshift Reddit API using NLP methods including TF-IDF, word2vec and sentiment analysis
• Built a recommendation system that is trained on words associated with positive experiences for a given drug. Employed Logistic regression and linear SVC for the classification
• Built an inter-active user interface with Flask, Bootstrap and AWS
  The repository for this project is shared