From Specification to Execution: AI Assisted Scientific Workflow Management

2025-05-18 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Expert, extended

Summary

An AI-assisted approach to scientific workflow management is presented, combining specification-driven generation, automated debugging, and distributed execution. This method introduces a structured specification phase, separating workflow intent, design, and implementation for validation prior to code generation. An LLM-based debugging agent diagnoses and resolves failures across multiple system layers. The system integrates Pegasus WMS with a Model Context Protocol (MCP) layer for unified interaction. Evaluated with a federated learning workflow for medical imaging, it generated and executed large-scale workflows with thousands of jobs, reduced debugging effort, and enabled non-expert users to construct expert-level designs. Claude Code, using the "pegasus-ai" plugin, produced a feature-complete workflow in 2 sessions for ~\$15-20, significantly outperforming OpenAI Codex and Kimi K2.6 in completeness and reproducibility, and reducing development time from an estimated 3-4 months to days.

Key takeaway

For MLOps Engineers designing complex scientific pipelines, adopting AI-assisted workflow management can drastically cut development time from months to days. You should prioritize systems that offer specification-driven generation and integrated LLM-based debugging, like the Claude Code with "pegasus-ai" plugin, to ensure reproducibility and autonomous fault recovery. This approach allows non-experts to achieve expert-level designs, freeing up specialized talent for optimization.

Key insights

AI-assisted scientific workflow management enhances reproducibility and reduces development effort through structured generation and autonomous debugging.

Principles

• Separate workflow intent, design, and implementation.
• Use structured specifications for validation.
• Ground LLMs with domain-specific skills/plugins.

Method

The proposed method involves a three-stage process: prompt/dataset description, structured specification generation, and code generation. It integrates an LLM-based debugging agent and Pegasus WMS with an MCP layer for distributed execution.

In practice

• Employ specification-driven workflow generation.
• Integrate LLM-based debugging agents for fault recovery.
• Utilize WMS like Pegasus for scalable execution.

Topics

• AI-assisted Workflow Management
• Scientific Workflows
• LLM-based Debugging
• Pegasus WMS
• Federated Learning
• Model Context Protocol

Code references

• opencode-ai/opencode
• pegasus-isi/claude-plugin-marketplace
• kthare10/fl-chest-workflow-kimi
• kthare10/fl-pegasus-workflow-gpt-5.4
• pegasus-isi/medical-imaging-fl-workflow

Best for: Machine Learning Engineer, Research Scientist, AI Scientist, AI Engineer, MLOps Engineer

Related on AIssential

• From Specification to Execution: AI Assisted Scientific Workflow Management — AI-assisted workflow management streamlines design, execution, and debugging through structured specification and LLM agents.
• Enterprises can now train custom AI models from production workflows — no ML team required — Enterprise AI application interactions generate valuable training data that most organizations currently fail to capture.
• LLM Eval Workflow: How to Build Reliable AI Quality Gates Without Vibes — Reliable LLM evaluation requires a structured workflow integrating diverse checks and continuous feedback, not just tools.
• How to Orchestrate AI Workflows — Pure AI workflows are fast to prototype but lack reliability; combine AI with deterministic orchestration for robust systems.
• How AI Is Reshaping Data Engineering: A Practical Guide With Examples — AI automates complex data engineering tasks, making engineers significantly more effective through self-healing pipelines.
• From Manual Document Processing to AI-Orchestrated Intelligence — An AI-orchestrated pipeline automates document processing using dual-model extraction, human oversight, and zero-code extensibility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.