Vision & Core Pain Points
The Initial Goal vs. The Real Problem
When I first started this project, the goal was simple: make EPA SWMM easier to install, run, reproduce, and check through an agentic workflow.
But as I kept developing it, I realized that real hydrological modelling is not only about running a model. A modelling workflow should also remember what was done, what assumptions were made, what failed, what was corrected, and whether the results can be trusted.
"To me, this is an important step from 'AI runs a SWMM model' toward 'AI helps build a reproducible, inspectable, and memory-informed modelling workflow.'"
The Three-Part Solution
That is why the current version of Agentic SWMM Workflow now includes three connected parts:
- Reproducible SWMM Execution: Run SWMM in a deterministic and inspectable way, generate run artifacts, and perform QA-oriented checks.
- Audit-Based Modelling Memory: Record provenance and produce Obsidian-compatible experiment notes to summarize repeated assumptions, QA issues, missing evidence, failure patterns, run-to-run differences, and useful modelling lessons.
- Controlled Skill Refinement: The memory layer uses these records to propose refinements to existing workflow skills.
Human Review & Safe AI
A key point is that this is not autonomous self-editing. While the memory layer can propose refinements to existing workflow skills, accepted changes still require human review and benchmark verification. This ensures that it is completely safe to use AI in real, professional projects.