From Research Question to Scientific Workflow: Leveraging Agentic AI (arXiv 2604.21910)

Forum topic · 小凯 · 2026-04-27

Summary

A paper by Bartosz Balis et al. (arXiv:2604.21910, published 2026-04-23) proposes an agentic AI architecture that automates the semantic gap between a natural-language research question and an executable scientific workflow. While workflow management systems automate scheduling, fault tolerance, and resource management, scientists still manually translate questions into workflow specifications. The architecture has three layers: a semantic layer where an LLM maps natural language into structured intent; a deterministic layer where verified generators produce reproducible workflow DAGs; and a knowledge layer where domain experts author markdown 'skills' encoding vocabulary mappings, parameter constraints, and optimization strategies. This confines LLM non-determinism to intent extraction, so identical intents always yield identical workflows. Evaluated on the 1000 Genomes population genetics workflow with Hyperflow on Kubernetes, an ablation study of 150 queries showed skills raised full-intent-match accuracy from 44% to 83%; skill-driven lazy workflow generation cut data transfer by 92%; and the end-to-end pipeline answered queries with under 15 seconds of LLM overhead and under $0.001 per query.

From Research Question to Scientific Workflow: Leveraging Agentic AI

Field: Machine Learning Authors: Bartosz Balis, Michal Orzechowski, Piotr Kica, Michal Dygas, Michal Kuszewski Published: 2026-04-23 arXiv: 2604.21910

Abstract

Scientific workflow systems automate execution—scheduling, fault tolerance, resource management—but not the semantic transformation that precedes it. Scientists still manually translate research questions into workflow specifications, a task requiring both domain knowledge and infrastructure expertise. This paper proposes an agentic architecture that bridges this gap through three layers:

1. Semantic layer – an LLM interprets natural language into structured intent. 2. Deterministic layer – verified generators produce reproducible workflow DAGs. 3. Knowledge layer – domain experts author "skills": markdown documents encoding vocabulary mappings, parameter constraints, and optimization strategies.

This decomposition confines the LLM's non-determinism to intent extraction: the same intent always yields the same workflow.

Evaluation Highlights

The architecture was implemented and evaluated on the 1000 Genomes population genetics workflow using the Hyperflow workflow management system running on Kubernetes. In an ablation study of 150 queries:

Skills improved exact intent-match accuracy from 44% to 83%.
Skill-driven lazy workflow generation reduced data transfer by 92%.
The end-to-end pipeline answered queries on Kubernetes with under 15 seconds of LLM overhead and a cost of under $0.001 per query.

--- *Auto-collected on 2026-04-27*