From Research Question to Scientific Workflow: Leveraging Agentic AI for Science Automation

TL;DR

This paper introduces an agentic AI architecture that automates the semantic translation of research questions into scientific workflows, improving accuracy and efficiency in workflow generation.

Contribution

The proposed multi-layered architecture combines language models, validation, and domain-specific skills to automate and improve scientific workflow creation.

Findings

Skills increase intent accuracy from 44% to 83%.

Skill-driven workflow generation reduces data transfer by 92%.

Pipeline completes queries with low latency and cost.

Abstract

Scientific workflow systems automate execution -- scheduling, fault tolerance, resource management -- but not the semantic translation that precedes it. Scientists still manually convert research questions into workflow specifications, a task requiring both domain knowledge and infrastructure expertise. We propose an agentic architecture that closes this gap through three layers: an LLM interprets natural language into structured intents (semantic layer); validated generators produce reproducible workflow DAGs (deterministic layer); and domain experts author ``Skills'': markdown documents encoding vocabulary mappings, parameter constraints, and optimization strategies (knowledge layer). This decomposition confines LLM non-determinism to intent extraction: identical intents always yield identical workflows. We implement and evaluate the architecture on the 1000 Genomes population genetics workflow and Hyperflow WMS running on Kubernetes. In an ablation study on 150 queries, Skills raise full-match intent accuracy from 44% to 83%; skill-driven deferred workflow generation reduces data transfer by 92\%; and the end-to-end pipeline completes queries on Kubernetes with LLM overhead below 15 seconds and cost under $0.001 per query.