The inspiration for this application was decentralized active marketing by word of mouth, since that is the best method of advertising delivering the most trust among brand names. When a close friend mentions a company that they truly support, companies want to be able to support those people. It is a quick way to earn small amounts of money through regular conversation. It encourages people to talk to each-other and spread the word of mouth about corporations who are doing things right.
What it does
This is a web-app that uses the microphone on your mobile device like "Google Assistant" or "Siri" to determine when someone mentions a brand that they support in conversation. The brand receives a great referral by word of mouth, which is the most psychologically successful marketing tactic, while being able to reward their long lasting patrons with either points that they can use in-store or additional discounts.
How we built it
This application is built on a Node.JS backend webserver hosted by heroku.com, with an Angular JS front end, and a custom UI built by Isaiah and myself. It uses a public script, annyang.js, to connect with google's voice servicing api, just like google's own assistant would use, and return a string of predicted phrases used during a conversation that it hears. When a string is matched to the brand that has been chosen by the user to be advertised, the application notes a successful match and is able to credit the user with points towards rewards for their word of mouth marketing. These points are stored in a MLab database on our webhosting and can later be redeemed and removed from the specific user account.
Challenges we ran into
Integrating the voice api that Google provides was probably the most difficult task of this project. This was mostly because it is such a powerful api that is able to return many different sets of data for a single query. For example, it can return full sentences of predictions that are most commonly used in terms of the machine learning that has been researched at Google. In other words, it can predict what you said even if it did not hear certain words, which makes understanding conversations quite easy for the software.
At first the oversight essentially was that the api was continuously returning multiple predictions, but once we were able to parse all of the data quickly enough, the point increment system was able to work.
Accomplishments that we're proud of
We started from scratch and were able to present a working demo of an angular web-application that can be run on any html5 capable machine, including smartphones that were used for the demo. Integrating with a high-level Google product was a great success as well.
What we learned
Voice to text api's are extremely powerful, but are extremely difficult to master since they have documentation that would exceed our time here if we were to read it. Connecting and managing voice requests on wireless platforms are extremely fragile and delicate. Who would have known vibrations caused by human vocal chords could be such a tricky concept to capture?!
What's next for Word-Of-Mouth
We have an entire list of extra features that could have been added as stretch goals if we had the time to do so:
- user login/universal web storage/single sign on authentication along with 2Factor Authentication
- creating a simple brand sign-up form for thousands of companies to sign on to advertise
- creating a better method to cash-out or transfer earned credits
- a smart time-out algorithm to stop potential abuse of the system
- using additional machine learning to understand sentiment and context in which the brand is mentioned (positive/negative)
- creating a direct link to provide feedback to brands on this application
- having a dynamically changing UI based on logo colors of chosen brand and/or device that the web-app is being used on