System / Architecture
A resilient decision-intelligence pipeline that combines custom machine learning, manufacturing analytics, and generative AI.
Orders, lines, quality, downtime
Validation and feature preparation
Trained tabular delay classifier
Optional Core ML service or web fallback
Evidence-based explanation
Corrective interventions and impact
Human validation and execution
A tabular delay classifier is trained and evaluated separately with Apple Create ML. The public web app uses deterministic TypeScript risk prediction unless an optional, separately hosted Core ML HTTP service is configured.
Gemini receives compact, relevant factory context and converts predictions and metrics into explanations and prioritized actions. A data-aware local response engine keeps the experience demoable whenever Gemini is unavailable.
The system goes beyond forecasting: it ranks the causes of risk, recommends capacity, staffing, material, maintenance, and quality interventions, and estimates the resulting risk reduction.
The production ML dataset contains up to 100,000 synthetic manufacturing observations, with stratified training, validation, and test splits. The dashboard uses a compact TypeScript demo dataset so it starts instantly and remains independent of CSV generation.