Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Opus: A Large Work Model for Complex Workflow Generation

Théo Fagnoni, Bellinda Mesbah, Mahsun Altin, Phillip Kingston

TL;DR

Opus tackles the challenge of generating and optimizing complex, industry-grade Workflows by integrating an explicit Workflow Intention with a Work Knowledge Graph to drive a fine-tuned Large Work Model that produces DAGs of Tasks and Instructions. The framework couples intention-driven generation with graph-based knowledge retrieval, then optimizes candidate workflows via a concrete cost model and a modified shortest-path algorithm, achieving auditable, semantically aligned workflows. In a Medical Coding use case, Opus Alpha 1 Large and Small outperform state-of-the-art LLMs by substantial margins across semantic and structural metrics, demonstrating the value of knowledge injection and intention formalization for regulated domains. The approach offers practical implications for cost reduction and quality improvement in BPO by enabling context-aware, executable workflows that integrate human expert reviews when needed, with a pathway toward scalable, production-grade deployment.

Abstract

This paper introduces Opus, a novel framework for generating and optimizing Workflows tailored to complex Business Process Outsourcing (BPO) use cases, focusing on cost reduction and quality enhancement while adhering to established industry processes and operational constraints. Our approach generates executable Workflows from Intention, defined as the alignment of Client Input, Client Output, and Process Context. These Workflows are represented as Directed Acyclic Graphs (DAGs), with nodes as Tasks consisting of sequences of executable Instructions, including tools and human expert reviews. We adopt a two-phase methodology: Workflow Generation and Workflow Optimization. In the Generation phase, Workflows are generated using a Large Work Model (LWM) informed by a Work Knowledge Graph (WKG) that encodes domain-specific procedural and operational knowledge. In the Optimization phase, Workflows are transformed into Workflow Graphs (WFGs), where optimal Workflows are determined through path optimization. Our experiments demonstrate that state-of-the-art Large Language Models (LLMs) face challenges in reliably retrieving detailed process data as well as generating industry-compliant workflows. The key contributions of this paper include integrating a Work Knowledge Graph (WKG) into a Large Work Model (LWM) to enable the generation of context-aware, semantically aligned, structured and auditable Workflows. It further introduces a two-phase approach that combines Workflow Generation from Intention with graph-based Workflow Optimization. Finally, we present Opus Alpha 1 Large and Opus Alpha 1 Small that outperform state-of-the-art LLMs by 38% and 29% respectively in Workflow Generation for a Medical Coding use case.

Opus: A Large Work Model for Complex Workflow Generation

TL;DR

Opus tackles the challenge of generating and optimizing complex, industry-grade Workflows by integrating an explicit Workflow Intention with a Work Knowledge Graph to drive a fine-tuned Large Work Model that produces DAGs of Tasks and Instructions. The framework couples intention-driven generation with graph-based knowledge retrieval, then optimizes candidate workflows via a concrete cost model and a modified shortest-path algorithm, achieving auditable, semantically aligned workflows. In a Medical Coding use case, Opus Alpha 1 Large and Small outperform state-of-the-art LLMs by substantial margins across semantic and structural metrics, demonstrating the value of knowledge injection and intention formalization for regulated domains. The approach offers practical implications for cost reduction and quality improvement in BPO by enabling context-aware, executable workflows that integrate human expert reviews when needed, with a pathway toward scalable, production-grade deployment.

Abstract

This paper introduces Opus, a novel framework for generating and optimizing Workflows tailored to complex Business Process Outsourcing (BPO) use cases, focusing on cost reduction and quality enhancement while adhering to established industry processes and operational constraints. Our approach generates executable Workflows from Intention, defined as the alignment of Client Input, Client Output, and Process Context. These Workflows are represented as Directed Acyclic Graphs (DAGs), with nodes as Tasks consisting of sequences of executable Instructions, including tools and human expert reviews. We adopt a two-phase methodology: Workflow Generation and Workflow Optimization. In the Generation phase, Workflows are generated using a Large Work Model (LWM) informed by a Work Knowledge Graph (WKG) that encodes domain-specific procedural and operational knowledge. In the Optimization phase, Workflows are transformed into Workflow Graphs (WFGs), where optimal Workflows are determined through path optimization. Our experiments demonstrate that state-of-the-art Large Language Models (LLMs) face challenges in reliably retrieving detailed process data as well as generating industry-compliant workflows. The key contributions of this paper include integrating a Work Knowledge Graph (WKG) into a Large Work Model (LWM) to enable the generation of context-aware, semantically aligned, structured and auditable Workflows. It further introduces a two-phase approach that combines Workflow Generation from Intention with graph-based Workflow Optimization. Finally, we present Opus Alpha 1 Large and Opus Alpha 1 Small that outperform state-of-the-art LLMs by 38% and 29% respectively in Workflow Generation for a Medical Coding use case.

Paper Structure

This paper contains 46 sections, 13 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Definitions
  3. Background
  4. Large Language Models for Workflow Generation
  5. Intention Capture
  6. Task layer in Workflow Generation
  7. Knowledge Graph-Informed LLM
  8. Workflow Optimization
  9. System Architecture
  10. Workflow Generation and Optimization
  11. Work Knowledge Graph (WKG)
  12. Preliminaries
  13. Nodes $N_{\text{WKG}}$: Tasks
  14. Edges $E_{\text{WKG}}$: Historical Workflow Implementations
  15. Intention
  16. ...and 31 more sections

Figures (4)

  • Figure 1: Workflow Generation and Optimization
  • Figure 2: Comparison of Opus Alpha 1 Large and Opus Alpha 1 Small to the top performing state-of-the-art LLMs across the metrics on the Medical Coding use case.
  • Figure 3: Medical Coding Opus Reference Workflow — simplified view
  • Figure 4: State-of-the-art LLMs performance across the metrics on the Medical Coding use case.