RAS Commander: When Large Language Models Meet Hydraulic Engineering Automation — A New Paradigm for LLM Forward Engineering Practice

RAS Commander is a Python library for automating HEC-RAS hydraulic model operations. It demonstrates how to use large language models (LLMs) to accelerate engineering software development while upholding the responsibilities of professional engineers. The project proposes the "LLM Forward" methodology, whose core principles include prioritization of professional responsibility, human-AI collaboration, and multi-level verifiability, providing a new paradigm for AI collaboration in the engineering field.

RAS Commander originated from the "AI Tools for Modelling Innovation" course at an Australian water school, evolving from William Katzenmeyer's initial teaching demonstration project. In less than two years, the CLB Engineering team built the most robust open-source automation solution for HEC-RAS and HEC-HMS. The project name "Commander" comes from the concept "Command Line is All You Need", embodying the concise philosophy of command-line automation.

Core Features

1. Project file operations: Read/modify HEC-RAS project files (.prj, .geo, etc.) and batch process model files
2. HDF data access: Conveniently read HEC-RAS 6.x HDF5 results
3. Simulation execution automation: Programatically start the computation engine, monitor progress, and process results
4. Test-driven development: Use official sample projects to ensure code reliability

Technical Details

• Architecture design: Replace the HECRASController COM32 interface with a Python library to support cross-platform operation
• Backward compatibility: Focus on adaptation to the HEC-RAS 6.x series
• Example-driven development: Provide complete runnable examples to optimize the LLM user experience

LLM Forward is a development philosophy pioneered by CLB Engineering, with four core principles:

1. Prioritization of Professional Responsibility: Public safety, ethics, and professional licensing are the top priorities; AI is only an auxiliary tool
2. LLM as Precursor, Not Priority: Accelerate engineering insights without replacing professional judgment
3. Multi-level Verifiability: Support GUI review, visual inspection, and code audit tracking
4. Human-AI Collaboration: Engineers supervise the entire process; AI-generated content requires manual review

The March 2026 update introduces the "fully agentized engineering experience" with core features:

• Human-AI collaboration mode: Display operation steps, maintain modeling logs, and generate reproducible deliverables
• Hierarchical knowledge structure: Long-term memory uses progressive disclosure, solidifying LLM-generated code into deterministic workflows
• File-based memory system: Support task planning across conversations/sub-agents

Typical application scenarios: Import external data, QAQC review, verify project configuration, AEP modeling, etc.

• Claude Code Agent: Built-in cognitive infrastructure for Anthropic Claude; the .claude/ directory contains agents, skills, and rules
• Codex Support: Read the AGENTS.md hierarchy and bridge shared skills via .agents/skills/
• Community Resources: ReadTheDocs documentation, ASFPM presentations, deprecated ChatGPT assistant (CLI agent recommended)

1. Efficiency Improvement: Automate repetitive tasks, reducing time by over 80%
2. Quality Assurance: Programmatic QAQC is more consistent; code audits meet documentation requirements
3. Knowledge Inheritance: Agent skills and rules preserve expert experience
4. Industry Demonstration: The open-source project shows a feasible path for AI collaboration in highly regulated fields

Future Directions

• Expand data format support
• Enhance visualization capabilities
• Improve documentation and tutorials
• Welcome community contributions

Conclusion

RAS Commander proves that professional responsibility and technological innovation can be balanced. The LLM Forward methodology provides a reference for various fields: Let AI be a tool to enhance human capabilities, not a shortcut to replace judgment.