Paper Contents
Abstract
This research introduces an AI-powered smart retail management system built on the Retrieval-Augmented Genera- tion (RAG) framework to enable dynamic sales forecasting. By integrating a Large Language Model (LLM) with Snowflakes vector database, the system ensures efficient data retrieval, while LlamaIndex enhances semantic extraction from structured tables. The methodology follows a structured pipeline: data collection from Kaggle, preprocessing to handle missing values and stan- dardize data, and loading the refined dataset into Snowflake. Using cosine similarity-based search, the system retrieves relevant sales insights and market trends to produce accurate, data-driven forecasts. To assess performance, the system was evaluated based on key metrics: Mean Absolute Error (MAE) of 95.7%, Root Mean Square Error (RMSE) of 94.3%, R Score of 87%, and F1-Score of 94%. These results highlight the models ability to minimize prediction errors while maintaining a strong balance between precision and recall. The combination of Snowflakes fast, scalable retrieval capabilities and LlamaIndexs advanced semantic understanding further enhances forecasting accuracy and inventory management. Beyond improving sales strategies, this hybrid approach also boosts operational efficiency. Future enhancements will focus on incorporating external data sources, refining embeddings, and exploring advanced hybrid models to further improve predictive accuracy and adaptability to shifting market trends.
Copyright
Copyright © 2025 Vanam Prabhas. This is an open access article distributed under the Creative Commons Attribution License.
