Adaptive FFDC Forex Assistant Powered by AI

Two Applications of Machine Learning with Supervised vs Unsupervised Learning Algorithms against an FFDC data powered Exchange Rate Portal

Comment

Inspiration

Use the powerful Machine Learning technologies to add up new, interesting features to applications that are getting powered by FFDC financial data

What it does

It applies two ML technologies (one supervised learning, the other one unsupervised learning) for two different purposes in the same application: 1) it automatically configure the component structure of an FFDC Exchange Rate Assistant based on the traffic and backend systems loading by using a model generated with 2-layer deep, 10-neurons per layer convolutional neural network (CNN). The model needs retraining and updates at regular intervals, and is persisted on the local server disk for better performance. The UI configuration commands are based on month, day and hour and they are provided through REST API 2) for the user who intends to buy currencies and sees the current exchange rate provided by FFDC API, an unsupervised regression algorithm (based on AutoRegression Integrated Moving Averages or ARIMA) provides exchange rates forecasts for the selected two currencies for the next N-day prediction interval (with N configurable)

How we built it

Tarjinder and I have used a stack of technologies that include FFDC REST API, Angular 8 (TypeScript), Python 3.5.6 and 3.7 (with Keras and Tensorflow 2.0), .NET 4.6.1

Challenges we ran into

One of the important challenges was related to the evolving versions of Python interpreters and the inability of some packages to keep up.

Accomplishments that we're proud of

This was a fun project and we are really proud of how it came up: cool changing interface driven by ML, decent performance, ability to be configured and evolve to a real solution

What we learned

We have learned that the new technologies, centered around machine learning (and its subspace of AI) are becoming widely available, easy to use and integrate and it takes only a grain of creativity to boost up the existing Finastra applications with new, cool, and highly marketable features

What's next for Adaptive FFDC Forex Assistant Powered by AI

We hope that this POC will attract the attention of Finastra technical and business stakeholders and we might get involved in activities that could lead to a fully fledged solution

Share this project:

Updates