Deep Learning models are used to predict patterns in complex sequences by learning from data, rather than relying on simple rules or algorithms. The models work by constantly iterating over the training data, making predictions and adjusting their parameters until the error between the prediction and ground truth is minimal. In conversations, using conversational data is challenging due to its rarity, but ensembling multiple Deep Learning models can provide exponential payoffs. This approach allows each model to learn specific things from different data sets, and combining them results in a better overall model. While even ensemble models can fail to spot simple patterns, additional techniques such as standard rule-based approaches can be used to refine the results.