It's not enough to build a trading bot that connects and operates on an exchange. The bot must also operate with sound algorithms that reflect the current market. For this reason, much of our work is dedicated towards building a solid infrastructure that allows for accurate simulation and testing.
Our continuous-build infrastructure periodically gathers market data and utilizes such data for regression testing. Regression testing automatically keeps track of the performance of existing algorithms and allows us to decide if some models should be revised.
We utilize Machine Learning techniques both while developing a reference algorithm to simplify the selection of fundamental indicators and relative parameters, and at run-time, where the bot constantly runs several internal simulations of many possible outcomes and selects the best performers based on continuously runs simulations of many possible strategies and selects the one that is best suited for the current market.
Our run-time is built in C++ from the ground up. This, added to our experience with high-performance applications, means that we can have the best performance possible both in terms of connectivity and in terms of number crunching to be performed in real-time.