Nestle

Solution

Nestle R&D worked with IBM Research to build a generative AI system for sustainable packaging discovery. The implementation starts with ingesting material knowledge from public and proprietary documents, which the team used to construct a knowledge base of known packaging materials. That curated corpus was then used to fine-tune a fit-for-purpose chemical language model so it could learn molecular structure representations relevant to barrier-material design. On top of that learned representation, the teams applied IBM Research's regression transformer to model correlations between structural molecular features and physical-chemical properties. The resulting model serves packaging researchers by proposing entirely new high-barrier materials that can protect sensitive products from moisture, temperature swings, and oxygen, helping Nestle evaluate future options against cost, recyclability, and functional packaging requirements.

Data flow

Public and proprietary documents are ingested to build a knowledge base of known materials, which is curated into a corpus for fine-tuning a chemical language model. Its learned molecular representations are used with a regression transformer that predicts structure-property relationships and proposes new high-barrier packaging materials for Nestle researchers to assess for cost, recyclability, and functional fit.