Where The Idea Came From
The U.S. government lost a track of 1475 immigrant children put in their custody. While trying to solve this issue, the team decided to expand the data parameters to include all children separated from their parents due to extreme circumstances like natural disasters, migration (or refuge), human trafficking.
The children who are lost, are of very young age and don't really remember their place of origin, date of birth to fill these forms. We aim to create a platform which asks the kids and their parents to enter unique things they remember about their family members, like tattoos, height, favourite tv show .... and our matching algorithm takes these parameters to uniquely identify the families.
Having so many features, created an issue of open-ended data points. To solve this issue, we are providing the users with the flexibility to enter as many features as they want, and not limiting the data points and applying Elastic Search and Machine Learning on these data points to match profiles.
How The Front End Theoretically Flows
A video plays in lieu of a text question and the child has the option to click on picture in lieu of text. Alt text will be the data set which is matched.
What It Does
It takes user defined responses and stores in a No-SQL based ElasticSearch framework. Using Machine Learning, it finds the best family members matches for the missing children/parents. It also stores masked personal responses in ElasticSearch and reveals email-only option to the user.
How We Built It
We built it using ReactJS on the Front-end, Node.js for middle tier, ElasticSearch for storage and quick-retrieval, Microsoft Azure as cloud platform, GitHub for version control, as well as Youtube and Google Suite for converting user responses into data.
Challenges we ran into
At first we faced problem of dealing with a large set of data points. We spent sometime to understand which data-model would best fit in to address our requirement of keeping responses user-defined. We also wanted to keep it short and simple so that a user does not lose attention span.
Coding challenges we ran into
We decided to switch to using ElasticSearch mid-project instead of ML, because of the data points not being fixed. This required everything to be recoded. Some of us were not really familiar with React or ElasticSearch so we learned while doing it.
Accomplishments that we're proud of
Effective data matching algorithm implementation.
What we learned
We learned that all need is to be on the same page before we begin. Never give-up and keep pushing yourself till the last moment!
What's next for Happily Ever After
We want to use FaceAPIs to find similarity between several facial features of family members to improve our matches in case user defined responses are enough. We would also like to have a message platform as the first place for the users to contact their missing family members. We hope that we can license this software and sell it to law enforcement officials, firefighters, and mandatory reporters that can help a child in a need.