InvenBot: An AI-Powered Chatbot for Intelligent Inventory Management

Abstract

This paper presents InvenBot, an intelligent in-ventory management system that integrates machine learning–based demand forecasting with a multilin-gual conversational interface supporting both text and voice interaction. The system provides real-time inventory tracking, analytics, and decision support across diverse product categories and brands. A key component of InvenBot is a Natural Language Processing (NLP) based chatbot that supports inter-action in English and Hindi, enabling users to query and manage inventory data through natural language using either text or speech inputs. The chatbot employs an agent-based architecture with multiple specialized tools for tasks such as product search, stock monitoring, revenue analysis, and demand forecasting, achieving an overall intent detection accuracy of 95.4%. Voice inputs are con-verted to text using speech recognition interfaces and processed through the same NLP pipeline, ensuring consistent performance across interaction modes. For predictive analytics, the system utilizes an XGBoost regression model trained on historical sales data with 26 engineered time-series features, including lag variables, rolling statistics, trend indi-cators, and seasonal patterns. The model generates demand forecasts for up to 12 months ahead, sup-porting proactive inventory planning. Experimental results indicate a training R² score of 0.9819 and validation R² of 0.3836, with a Mean Absolute Error (MAE) of 30.89 units and Root Mean Square Error (RMSE) of 51.14 units on the validation set. To ensure robustness, the system incorporates fallback mechanisms based on moving averages in cases where model predictions are uncertain.