Blog

Train Once, Infer a Million Times: Bayesian Inference That's (Finally) Cheap at Query Time

MCMC re-runs from scratch on every new dataset. Amortized Bayesian inference pays the cost once -- then gives you a posterior in milliseconds, forever. Stefan Radev (BayesFlow creator) and I built an Agent Skill that makes it reliable. Come join us, your first trained amortizer is ~3 minutes away 😉

I Built an Agent That Refuses to Say "Causes"

Coding agents confidently say "X causes Y" without drawing a DAG, checking assumptions, or running refutation tests. I built an Agent Skill that won't let them. 100% eval pass rate vs 68% without -- the gap is in the reasoning, not the code.

A Researcher Stress-Tested My Bayesian Agent Skill. Here's What Broke

A researcher stress-tested my Bayesian Agent Skill on a real hierarchical model. It caught the workflow violations but missed a subtle prediction bug, had no sparsity priors, and lacked classification evaluation. Here's what I shipped to fix it

Bayesian Workflow Agent Skill for Coding Agents

Teach your AI coding agent proper Bayesian workflow in PyMC. This agent skill covers priors, model checking and diagnostics, and stops subtle statistical errors.

From Posterior to Profit: How Bayesian Decision Theory Turns Models Into Real Business Value