This document discusses approaches for time-series prediction using neural networks and analyzing customer order data from Instacart. It describes using word embedding models like word2vec to predict future purchases based on past order patterns and explore relationships between product purchases over time. It also emphasizes the importance of joining time-series data with metadata in a database to enable richer predictive queries and analysis. Time-series databases are presented as a solution for scaling to large volumes of IoT and sensor data and providing efficiencies over traditional databases by treating time as a first-class citizen.