Inspiration

Saving Lives One Tweet at a Time

What it does

An intelligent twitter bot that replies to medicine requests with actionable information (i.e. previous tweets with offers and pharmacies that have the medicine)

How I built it

Text Processing:

  1. text to lower
  2. Remove punctuation marks and numbers
  3. Replace medicine names for a keyword
  4. Remove stop words
  5. Apply Spanish stemmer to the words
  6. Create a vector representation of the text: size 100

Use NLTK and SKLEARN for the Tweet Classifier

Challenges I ran into

  1. Tweets with Images
  2. Lack of Data
  3. Integration challenges in GCP

Accomplishments that I'm proud of

This system replies in quasi-realtime to people that need medicine now!

What I learned

Building bots that respond in real-time is hard

What's next for MediTweet

  • Make the system more scalable and robust
  • Process images within tweets
  • Improve Classifier

Credits

  • Special Credits to the folks from BUSCAMED (github)
  • Leo Urbina, for helping use set up caching on the edge.
Share this project:
×

Updates