Inspiration
After listening to the challenges and datasets provided by this Hackathon's sponsor companies, we favored Grenke's challenge: The use cases described were broad enough to include lots of own ideas, what our creative and motivated team members value a lot, and those ideas lead to us not only using the data provided, but also to collect and use data retrieved from several external sources. In addition, we were able to work very close with people from Grenke, to be able to truly understand the problem, and develop an idea of the real impact of our product for the company.
What it does
Our idea was to improve and simplify Grenke's current process to decide whether a leasing request is accepted or not. To do so, we trained some machine learning models based on the data provided by Grenke. These models predict, based on data features like contract details and company credibility, how likely it is that Grenke accepts a leasing request. Additionally, in order to simplify this process even further, we included data from external sources that have an impact on the accepting decision. This additional data includes information about products included in the leasing request retrieved from online shopping APIs, such as the product's price, as well as further details about the company requesting the lease. All data collected and calculated for a specific leasing request is then shown on a website.
How we built it
Our product consists out of multiple separate services: The AI Server, the Frontend Server and the ML Service. The AI Server receives leasing data, collects and calculates additional data for a leasing request, stores this data and provides an endpoint for the Frontend Service to retrieve this data. It is written in Python 3, the frameworks used to run as a webserver with REST API are Django and djangorestframework. To be able to search external websites, we used the library beautifulsoup4. The Frontend Server, being responsible to serve a website to the actual user, is written in Angular. It is managed in NodeJS and we used scss to style the resulting page, and make it convenient for the viewer. Our machine learning models are trained using Microsoft Azure, with which we also created an endpoint that is called by the AI Server to calculate the machine learning based leasing score for a request. The services communicate via APIs to ensure each component can be used separatly and thus are interchangable with services provided by Grenke.
Challenges we ran into
As we worked very close with Grenke's mentor when starting with this challenge, and the dataset we got in the end was not a "standard one", but selected together with the mentor, the time we spent diving deep into the problem and the data available took us longer than expected. This ment that we were not able to start training machine learning models as soon as the other teams. On the positive site, this extensive process enabled us to gain a deep understanding of the data and, thanks to this mentor, a detailed discussion of the actual work flow at Grenke.
Accomplishments that we're proud of / What we learned
An important goal all of our team members had for this Hackathon was to experiment with new technologies and learn something new - for most of us it was the first time working with Python and Angular, and we are proud that in the end our product turned out quite ok ;). Also, as all of us are students, we were quite excited about being able to use real world data and we are very pleased with the outcome.
What's next for DEEPFRIED:AI
Get some sleep.
Built With
- angular.js
- azure
- django
- json
- machine-learning
- microservice
- python
- rest
- scss
- sql
- typescript
Log in or sign up for Devpost to join the conversation.