The engineers at Funding Societies are building some really cool stuff on a regular basis. We are pushing out new features every week and new products (external and internal) every month. In fact Feburary ‘17 was internally known as the launch month! As the name suggests we launched a bunch of cool new products and features this month. Amidst all this, our engineers often get so busy meeting deadlines and shipping code that all the fun ideas we want to build keep piling up on our idea stack. On a random weekend in Feb, we took an opportunity to dedicate 36 hours out of the daily grind and start clearing our stack.
In short 11 engineers, 2 days and 8 projects. We decided on ideas and teams before Friday morning. Since there were only 11 of us we constrained team size to max 2 people, also accommodating quite a few who chose to be lone rangers.
The theme was simple: build something that helps our valuable users or ourselves. That’s it. No other restrictions. Out of the 8 projects that were built, we are sharing about 3 publicly in this post. We can’t share everything here (competition, compliance, and other mumbo jumbo). Join us to find out more about our efforts or even better, your project could be featured here for the next hackathon!
This project allows you to take a selfie with your photo ID and be verified real-time. If the photo on the ID is the same as the person holding the ID - Verification Success.
The face-matching project is built upon the open source project - facenet. On the mobile app, two faces will be detected and extracted from the photo, resized and sent over to backend api. On the backend api, our algorithm will first find out the landmarks of faces and then align them with rotation and translation. Next, a pretrained inception net v3 (by Google) is used to extract up to 1900 features from the two faces. As the features are represented as decimal numbers, we compute the sum of square distances between two feature vectors. By evaluating against LFW dataset, we are able to arrive at the distribution of distances for same and/or different person(s). These two sample distributions have distinctive sample mean, with few overlaps, as shown in figure below. Based on the distribution, a probability of two faces being the same person is be calculated.
2. MiYu bot
There has been a lot of excitement around chatbots in the media recently. Yet most bots in the market provide basic functionality. Our aim was to kickstart development of a multi-feature investor bot using which you can securely login, get new investment and repayment updates, and actually invest in crowdfunding rounds via a chat interface right in your Facebook Messenger. Later, we added a simpler workflow for borrowers to reach out and apply for loans as well. We have incorporated some NLP features to make the process easier. The bot is not self-aware yet, but does make you laugh every now and then. It has been christened MiYu after our adorable Customer Experience Specialist Yumi, whom you may spoken to on Live Chat at our site. In fact, one of our interviewees thought Yumi was a bot! This is how it all started. To clarify - Yumi is real person, MiYu is a bot :)
|Investor Flow||Investor Flow(cont.)|
|Borrower Flow||Chit Chatter|
3. Bank Journal
As a digital lending company, we examine financial data such as bank statements of our users. To make loan application easier for our borrowers, we allow login to their accounts via our Mobile App (check out FS Bolt!), for easy and secure sharing of bank statements with FS. A lot of backend work enabled a level of automation new to the industry. In a spirit of giving back, we have decided to share some of it with the developer community of South East Asia (SEA) so that folks can use these APIs to build their own versions of Mint for SEA.
During the hackathon we added ability to share your Bank Statements with the service by taking a photo of it (using Google’s Cloud Vision API). We plan to keep on shipping features to this tool as we stay true to our goal of giving back to and serving the developer community that we all collectively benefit from all the time.
We had a ton of fun building these products! If you would like to be part of the fun, join us! Send your resume to firstname.lastname@example.org!