Scalable Backend Architecture with Hybrid Machine Learning for MSME Operations Optimization: Demand Forecasting and Credit Risk Assessment

India’s micro, small and medium enterprises (MSMEs) contribute close to 30% of national GDP which is very significant, yet the vast majority still uses manual and inefficient paper ledger method to maintain their data. Because the margins are low in the MSMEs, the process of maintaining stock replenishment and credit record maintenance becomes crucial and due to the traditional methods these two activities are prone to inaccuracies. We present a cloud deployable intelligent platform designed for this context, using a Node.js / Express.js REST backend and MongoDB document database store with two dedicated machine learning models running on flask server. The first service is a two stage demand forecasting engine in which Facebook’s Prophet decomposes trend and multi period seasonality, with an XGBoost regressor that works on the systematic residuals left by that decomposition. It was tested on five stores from the Rossmann Store Sales benchmark, this hybrid model achieved a mean SMAPE based accuracy of 86.79% (peak 95.88%), beating Naive, ARIMA, Seasonal Naive and standalone Prophet; Diebold-Mariano tests confirm significance at p < 0.01. The second model is a credit default classifier that works on a schema built dedicatedly for the Indian udhaar lending culture, where festival period and salary week are key indicators maintained in every record by an automated ETL (Extract Transform Load) layer. Evaluated for 28,638 real loans from the Kaggle Credit Risk benchmark, XGBoost achieved ROC-AUC=0.8672 and a cross validated ROC-AUC score of 0.9431 ±0.0034, performing better than Logistic Regression, Decision Tree and Random Forest (McNemar test, p < 0.05).