Using Petri nets as a formal language for LLM-assisted development (github.com)

go-pflow is a new Go port of the pflow.xyz Petri‑net ecosystem that combines formal Petri‑net modeling with ODE simulation, process‑mining, real‑time predictive monitoring and a “Neural ODE‑ish” parameter‑fitting pipeline. The library exposes both an explicit and fluent Builder API, JSON‑LD import/export compatible with pflow.xyz, a CLI, SVG visualization and many packaged examples (SIR epidemics, game AI, Connect4, Sudoku, knapsack, hospital workflows). Under the hood nets are translated to mass‑action ODEs solved with adaptive solvers (Tsit5, RK45, implicit methods), include automatic steady‑state detection, caching and solver tuning (Dt, Dtmin/Dtmax, abstol/reltol). For the AI/ML community this is significant because it offers a reproducible, interpretable bridge between discrete process models and continuous differentiable dynamics: process discovery (Alpha/Heuristic miners), rate learning from event logs, sensitivity/gradient ranking, and learnable transition rates enable data‑driven calibration and Neural ODE‑style fitting. Real‑time predictive monitoring and SLA forecasting make it suitable for production operations and online ML-assisted controllers; reachability analysis, actor/state‑machine compilation and hypothesis evaluation enable hybrid modeling for games, optimization and decision making. The project is AI‑native with structured JSON outputs and assistant guidance (CLAUDE.md), making it easy to integrate into ML workflows, automated validation loops and research pipelines.