2019
cryptoAI

This deep learning software was developed to forecast the most profitable trading signal for cryptocurrencies at any given point in time. The predictions are generated using artificial neural networks (CNN & RNN/LSTM hybrid) that were trained on historic price data. The predicted trading signals can be used for simulating trading on a historic time span or used for real trading in real-time.
The neural network part of this software started out as my master thesis and is written in Python, Tensorflow and Keras. The software that carries out the trades in real-time at an actual crypto exchange is written in TypeScript using Node.js. Both parts communicate via WebSockets.

  • Neural network predictions
  • Data aggregation of historic crypto data
  • Feature extraction, dimensionality reduction, preprocessing
  • Arbitrary intervals (1sec, 30sec, 1min, ...)
  • Real-life trading
  • Backtesting
  • Considering trading fees
  • Quick prototyping of new neural networks