Case Study: Shell improves domain-specific LLM understanding with Weights & Biases

Shell boosts domain-specific understanding 30% with Weights & Biases

Shell, a major multinational oil and gas company, faced the challenge of utilizing its vast, siloed institutional knowledge, which was vulnerable to being lost. To create an AI assistant that truly understood its specialized domains, Shell's NLP research team partnered with vendor Weights & Biases to build a domain-adapted large language model.

The solution involved a complex pipeline for data processing and model tuning, heavily utilizing Weights & Biases for experiment logging, hyperparameter optimization with W&B Sweeps, and evaluation with W&B Weave. This resulted in a 30% improvement in domain-specific understanding and a 26% increase in accuracy compared to baseline models, creating a scalable system for continuous AI innovation.

Shell