Twitter analysis machine learning (ish?)

So, myself and some friends have decided to spend our spare time working on a start-up idea. A part of it depends on sentiment analysis of data (Mainly tweets). During our investigations we noticed that all the sentiment analysis apis currently out there are pretty poor. I am a front end developer, so machine learning is out of my grasp, but I'm convinced there there must be a better way.

My questions is.... If you were to implement a solution that could work out if people were social talking positively or negativity about a subject, whilst also identifying trends, what approach would you take/ what advice could you give?

Pretty big and open question, but any thoughts would be awesome!