Agent · Process optimisation & query
Icebreaker
Optimises batch weight-transfer accuracy in ice-cream production, and answers operations questions in plain English.
Icebreaker pairs a behavioural twin — a neural network that learns the operational characteristics of a specific production line — with an LLM reasoning agent that proposes parameter adjustments and explains them in terms an operator can verify. A second loop turns plain-English questions about production history into SQL, then back into insights — putting the data the line already generates within reach of every team member, not just the analyst.
Key features
- Behavioural twin: per-line neural model that predicts how parameter changes propagate to outcomes.
- Reasoning agent: human-readable recommendations with confidence ratings and rationale.
- Natural-language query of production history — no SQL required.
- Closed feedback loop: every operator decision deepens the model and the institutional record.
Status: Live deployment at Frosty Boy · access on request