The paper is a description of an artificial intelligence inventory application that will streamline the process of e-commerce. The proposed methodology combines the best preprocessing methods like the use of Robust Z-Score Normalization and Winsorization to deal with noisy and outlier-prone data to have better inputs to be used in training the models. To select the features, Minimum Redundancy Maximum Relevance (mRMR) is used to select the most informative variables at the expense of less complex computations. The predictive model consists of Hybrid Temporal Convolutional Network and Bi-LSTM to both account for the short-term fluctuation and the long-term demand trends in inventory. The system is implemented on the Databricks Lakehouse platform and has been made to be scalable, high-performance, and in real-time. The model is able to show considerable gains in forecasting, reduction of overstock and stockout, as well as inventory turnover and is a scalable, adaptive and efficient system of inventory optimization used in e-commerce. Findings have shown that AI-powered approaches have the potential to revolutionize the inventory management process that can have significant operational advantages to businesses functioning in contemporary e-commerce