StockSense AI: An Intelligent Inventory Forecasting and Explainable AI Platform for Small and Medium Enterprises

StockSense AI is an intelligent inventory forecasting and explainable AI platform designed specifically for e-commerce small and medium enterprises (SMEs). It addresses the key pain points of traditional inventory forecasting tools—complexity, difficulty in use, and the need for professional knowledge to interpret results. The platform integrates Prophet and LightGBM for time series prediction, provides model interpretability via SHAP values, and uses large language models (LLM) to convert data into direct, actionable business recommendations, enabling business owners without technical backgrounds to easily make data-driven inventory decisions.

For e-commerce SMEs, inventory management is a core challenge. Traditional tools output complex tables and charts that require professional data science knowledge to interpret, but most business owners lack this background and urgently need clear, actionable business recommendations (such as restocking timing, quantity, and reasons). StockSense AI was born to address this pain point, converting complex machine learning predictions into easy-to-understand natural language recommendations.

Integrated Prediction Engine

Uses a Prophet+LightGBM combination: Prophet captures seasonality and trends, while LightGBM handles non-linear relationships, generating high-precision demand forecasts for 7-30 days with confidence intervals. Zero-padding and calendar reindexing are used to avoid over-inflated predictions.

Intelligent Promotion Recommendations

Analyzes holidays, excess inventory thresholds, and weekly sales patterns to generate three types of recommendations: pre-holiday warm-up promotions, inventory clearance promotions, and weekend flash sales, displayed on a glassmorphism UI planning panel.

Explainable AI (XAI)

Uses SHAP TreeExplainer to identify key drivers of predictions (e.g., holidays, promotions, weekend effects), enhancing user trust and understanding of decision logic.

LLM-Driven Insights

Packages forecast data, SHAP factors, and business context and sends them to LLMs (supports Ollama/Groq) to generate actionable recommendations (e.g., stockout warnings, restocking suggestions).

Backend Architecture

FastAPI/Uvicorn (API routing and JWT authentication), Prophet+LightGBM (prediction engine), SHAP (interpretability), DuckDB+Pandas (data analysis), Ollama/Groq (LLM integration), FPDF (PDF weekly reports), SQLite+SHA-256 (secure storage).

Frontend Design

Glassmorphism style with dark mode support, dynamic visualization via Chart.js, designed to be 'understandable without a data science degree'—complex metrics are converted into intuitive visual elements and natural language.

StockSense AI brings multi-layered value to SMEs:

• Lower decision-making threshold: No need to hire data scientists or learn complex tools—directly get recommendations on 'what to do';
• Reduce inventory costs: Accurate forecasts avoid capital occupation and storage costs from overstocking;
• Prevent stockout losses: Timely warnings and restocking recommendations ensure hot-selling products are never out of stock;
• Optimize promotion ROI: Data-driven promotion recommendations give marketing campaigns clear goals and expected results.

Limitations

• Primarily targeted at e-commerce scenarios; adaptation to manufacturing, retail, etc., requires additional customization;
• Prediction accuracy depends on the quality of historical data; new stores or products with severe seasonal fluctuations need a longer data accumulation period.

Outlook

• Introduce reinforcement learning to optimize dynamic adjustment of promotion strategies;
• Integrate upstream supply chain data to achieve end-to-end collaboration.

StockSense AI is a successful example of AI implementation for SMEs. Instead of showing off technical complexity, it uses technology to lower the threshold of use. By combining machine learning, explainable AI, and LLM, it truly enables 'data to speak' and provides SMEs with intelligent decision-making capabilities that were previously only affordable for large enterprises.