PillID

• 

  The homepage

• 

Inspiration

The plethora of pills that I consume on a daily basis for my wide assortment of medical reasons.

What it does

It identifies pills using a neural network and displays aggregated information about interactions with alcohol, pregnancy, breastfeeding, and a wide assortment of other data. I only gathered information and photos on four medications as I did not have the time to build an automotive script for mass collection.

How we built it

With lots of caffeine. That's how. All jokes aside, I built this by creating a solid front-end then working my way to connect it back to its neural network core. I took it step-by-step, gradually connecting each individual component into a fluid system.

Challenges we ran into

I was unable to implement the optical character recognition algorithm to prepare this platform for being able to identify significantly more pills by cross-referencing their datasheets with the text on them. This decreases the validity of the results in a scaled-up version.

Accomplishments that we're proud of

This is my first project using a neural network and this is also my first time touching front-end development in over five years so I had to relearn everything, and I am proud of how my site tuned out in the face of that initial hurdle.

What we learned

I learned how to create a neural network for classification and how to implement it using a website as an interface.

What's next for PillID

Three things. 1) Moving the neural network script to a server instead of being executed on the client. 2) Adding another neural network that detects the imprints and text on a pill. 3) Vastly expand the number of medications that this software can identify by web scraping and using databases. This is an incredibly scalable project due to how much data is available online and how big this project can get by automating the data collection.

Built With

• azure
• bootstrap
• css3
• html5
• javascript
• jquery
• tensorflow