Distribird: AI-Powered Tool Automates Literature-Informed Priors for Bayesian Calibration

A new agentic web application called Distribird automates the construction of informative prior distributions for Bayesian model calibration, addressing a long-standing bottleneck in scientific modeling. The tool, described in a paper on arXiv (2608.11210), deploys a multi-agent pipeline that searches scientific literature, extracts parameter values, weights them by domain relevance, and fits probability distributions via AIC model selection. It is designed for process-based models with physically interpretable parameters, where uniform priors are commonly used due to the difficulty of manually building informative ones. Distribird provides clear reporting of evidence and confidence levels for each prior, and falls back to uninformative alternatives when literature is unavailable. The system aims to reduce the need for both domain and statistical expertise in prior specification, potentially accelerating research in fields relying on Bayesian calibration.