This document summarizes Qraft's Deep Asset Allocation engine, which uses deep learning to perform tactical asset allocation. It trains on macroeconomic and market data to extract features and learn an optimal allocation strategy. It aims to improve on traditional approaches by continuously learning from new data. The document outlines problems with deep learning like small data and presents Qraft's solutions, such as pretraining and uncertainty quantification. It validates the strategy's performance against benchmarks on test data, demonstrating higher returns with lower drawdowns.