We take a still of your face, and pool emotional/topic predictions from the Microsoft, Indico, and Clarifai APIs to spark a conversation with an AI. (with your news feed?)

Using Wikipedia as a training set and fictional news feed data as the test set, we sort the posts into the topics you might find interest or delight in based on your emotions, with help from gensim and Indico.

Rainier - Web development, APIs, ML, and deployment
Minah and Clarissa - This fun idea :), graphic design, creative direction, learned web development

• Microsoft Project Oxford
• Indico
• Clarifai
• gensim
• DigitalOcean
• Flask (Python)
• Bower for front-end dependencies (d3, bootstrap)

• Grabs the binary data from the WebRTC still grab attached to POST
• WebRTC was somewhat intuitive to use--unicode or ASCII byte-encodings not-so-much. Manipulates the unicode that Python gets from the POST request form dict and turns it into the appropriate ASCII byte-encoding, which is then base-64-decoded, and then piped into a random UUID-named .png file.
• As a hack, I used sshfs to mount the public_html directory of my UT Austin CS account address into my working dir, and sent the new .png files into that folder, chmoding per upload.
• This renders the image into a resource that's easily accessible by APIs. (Although this obviously won't scale, I only have 2 GB as an undergrad.)
• Finally, sends the URL via POST to the Microsoft Emotions API
• tl;dr I changed an image data-URI to a publicly available URL so it'd play better with some ML libraries that didn't have native Python clients, but did have RESTful APIs. And this took longer than anticipated.
• if you ask, changing to application/octet-stream in the HTTP header did not work for me... but I learned a lot more from this anyway! 😁

• One day I will get better at front-end! 🍻

• One day I will get better at ops! 🍣

• Extraction of related tweets from status update
• Visualization of topic modeling tweets with network graphs
• Store the data-URIs to a database
• Integration with a chatterbot whose training state is on a database (or at least keep chat history and bot training live for the session with something like Redis). Use long polling or websockets for smooth UX