Abstract
Valuing agroindustrial residues within the biorefinery framework promotes the bioeconomy by offering alternatives to traditional processes. Conducting early risk assessments is crucial for guiding decisions and identifying safe production methods, covering biomass handling, solvent transfer, reactive and separation processes, and operational conditions. Challenges include limited data and the manual effort to compute safety indices. This paper introduces an AI agent for autonomous estimation of the Inherent Safety Index from simulated data from two avocado biorefineries. It compares ChatGPT-based language models —Economy (GPT-4.1 mini), Balanced (GPT-5.1), and Advanced (o4-mini-deep-research)— against expert calculations. The advanced model achieved 92.9% accuracy in a simple biorefinery, though with higher costs. The balanced model reached 98.7% accuracy in a complex biorefinery, reducing latency and costs. These findings highlight the trade-offs among cost, accuracy, and speed, showing that auditable AI can enhance safety screening and support expert judgment.
