The defect predictor is one capability. The intelligence around it is the system. Here's how four distinct AI layers combine into a foundry operations co-pilot you can actually trust.
"Talks back."
Plain-English Q&A over your heat data. No SQL, no dashboards โ just ask.
โ LIVE in this demo"Always on."
Continuous anomaly detection flags drifting heats before they become defects.
โ LIVE in this demo"Decides for you."
Bayesian optimization finds the precise setpoint combination, no trial heats.
โ LIVE in this demo"Honest about uncertainty."
Every prediction ships with a confidence band. Wide bands = bring in a human.
โ LIVE in this demo"Talks back."
Every operator, plant head, and quality engineer in your foundry has questions about their data. Today, those questions take hours โ open SAP, export to Excel, build a pivot table, talk to the analyst, wait. FoundryOps Copilot answers them in seconds, in plain English, with the numbers cited.
Your team doesn't learn SQL. They don't learn dashboards. They just ask.
Stack: Anthropic Claude API + structured data context
User question โ backend assembles a relevant data summary from the heat database โ sends to Claude with a system prompt explaining the dataset โ returns a natural-language answer with cited figures.
"Always on."
Most foundry incidents aren't catastrophic โ they're slow drifts. A pattern wearing past its threshold. A furnace lining thinning. An operator's pour delay creeping up by 30 seconds a week. Humans don't notice. The AI does.
An unsupervised model continuously watches every heat for unusual combinations of parameters โ combinations that historically correlate with defects. When something looks off, it flags the heat before pour, not after rejection.
Stack: Isolation Forest (scikit-learn)
Trained on 5,000 historical heats. Each new heat scored on a 0โ1 anomaly scale. Anomalies above 0.7 trigger an alert. The explanation is generated by identifying the 2 parameters that contributed most to the anomaly score.
"Decides for you."
Prediction is useful. Action is valuable. The defect predictor tells you a heat is risky. The optimizer tells you exactly what to change โ and proves the new setpoints will work, before you pour.
Your senior metallurgist's tribal knowledge is "if porosity is up, drop moisture and increase pour temp by 10ยฐ." The AI finds the precise, multi-parameter optimum โ across 12 dimensions simultaneously โ in 3 seconds. No trial heats. No guessing.
Stack: scikit-optimize (Bayesian optimization with Gaussian Process surrogate)
"Honest about uncertainty."
The worst AI systems are confident when they shouldn't be. The best ones tell you when they don't know.
Every prediction in FoundryOps Copilot ships with a confidence interval. When the model has seen plenty of similar heats and the prediction is stable, the band is tight. When the heat is unusual โ extreme parameters, rare combinations โ the band widens and the model says so.
This is what separates an AI tool from AI snake oil.
Stack: XGBoost with prediction-interval estimation
For each prediction, we compute the entropy of the per-class probability vector. We combine that with the training-data density in the local neighbourhood of the prediction (KDE-based).
Wide intervals are not a model failure. They are the model's honesty.
A complete AI system isn't one capability. It's four working together.
FoundryOps Copilot is built around this principle: every layer reinforces every other layer. The Q&A surfaces insights the anomaly detector found. The optimizer respects the confidence intervals. The system tells you what it knows, what it suspects, and what it doesn't know.
That's the difference between AI you demo once and AI your team uses every shift.
The four layers above are live in this demo. Here's what we build next together.
Ready to see them work on your data?