Indusmind: An AI-Assisted Decision-Support Prototype for Operational Risk Visibility in Small-Scale Manufacturing

Small manufacturers must balance machine condition, order commitments, and material stock, yet the relevant records often sit in separate spreadsheets or tools.IndusMind is a web prototype that brings machine, production-order, and inventory records into one interface and raises alerts from telemetry values, order status, and stock thresholds.It uses a React and Vite frontend, a Node.js and Express API, Mongoose models with an in-memory demonstration fallback, and a scikit-learn logistic-regression model for machine-risk scoring.The risk view blends model scores with fixed rules, whereas the delayed-order and low-stock alerts are purely rule-based.The natural-language "copilot" matches keywords and returns templates; it does not call a generative language model.The classifier was trained on 1,200 synthetic telemetry records.On a stratified 80:20 holdout it reached an accuracy of 0.7667, precision of 0.4684, recall of 0.7255, F1-score of 0.5692, and area under the receiver operating characteristic curve of 0.8120, whereas a majority-class baseline reached an accuracy of 0.7875.Because both the records and the labels are synthetic, these figures describe a development experiment and say nothing about factory performance.The contribution of this work is an integrated prototype whose alerts and data flow can be inspected.Field validation, calibration on real data, and usability testing remain necessary before any operational benefit can be claimed.