Inspiration

With recent concerns over popular social media influencers posting inappropriate and questionable content for their followers to see, it has become vastly more difficult for parents to monitor the content that their children consume. With this in mind, we set out to build a platform that would enable parents to easily keep a_ watchful eye _on the social media content their children are exposed too.

What it does

Watchful AI is a web app where parents can enter their child's Instagram username in order to monitor their online consumption. Watchful AI extracts a list of the users they are following and then uses machine learning on each user's most recent posts and classifies the image as either safe or unsafe. Watchful AI can also handle video content and provide a rating of how safe the video is.

How we built it

We knew we wanted to use stdlib to run the analysis of the images so we started there. We started by creating a stdlib function for analysing an image and calculating a "safety" score (how "good" the image was). Then we began working on the node.js back to get the Instagram API to work. We discovered that the API did not offer the ability to find who a certain user was following so we had to use another platform to extra that data. We used Parsehub to scrape the usernames of the followers and then we used the Instagram API to get some of the last posts. The API limits us to 20 posts max and we can only access the posts from people in our "sandbox". So this hack currently only works with a select number of accounts, we would need to submit the app for review before it can be applied to any account. Once the backend was interacting with the API and stdlib functions correctly we added a front-end built on Bootstrap and jQuery. Overall the hack serves as a great proof of concept!

Challenges we ran into

  • The Instagram API is incredibly limited and made it a challenge to accomplish anything.
  • The Instagram API doesn't allow us to retrieve a user's following list, thus we had to scrape the web page to acquire that data
  • We ran out of stdlib credits
  • The stdlib timeouts were a problem and forced us to restructure our calls.

Accomplishments that we're proud of

We managed to get a mostly working web app even though we both had very limited to no experience with web design and javascript. We are also quite pleased with the accuracy of the model we trained.

What we learned

  • How to use stdlib
  • How to use Clarifai
  • Instagram sucks

What's next for Watchful AI

We would like to implement similar functionality for other social media sites such as Twitter, Facebook and Youtube. As well as improving the accuracy of the prediction model.

Built With

Share this project:
×

Updates