Selected work

Representative Papers

Opus: A Quantitative Framework for Workflow Evaluation

Nov 06, 2025

This paper addresses the absence of a unified quality quantification standard for automated workflows. We propose Opus, a novel evaluation framework that jointly models four orthogonal dimensions—correctness (success rate), reliability (structural consistency and information hygiene), efficiency (resource consumption), and value (output gain)—integrating reward mechanisms with normative penalties. Grounded in expected utility theory, structured coupling metrics, observability analysis, and information entropy detection, Opus constructs a probabilistic scoring model. The framework enables automatic workflow scoring, cross-process comparison, and multi-objective optimization, and can be embedded into reinforcement learning loops to support end-to-end workflow discovery and iterative refinement. Experimental results demonstrate that Opus significantly improves both efficiency and reliability of automation systems and accurately identifies Pareto-optimal workflows in complex scenarios.

0 citationsRead paper

Opus: A Prompt Intention Framework for Complex Workflow Generation

Jul 15, 2025

To address the insufficient quality and reliability of LLM-generated workflows for complex, multi-intent user queries, this paper proposes Opus, a prompt-based intent framework that introduces a reproducible and customizable intent-capture layer between natural-language queries and workflow generation. Opus decomposes mixed intents into structured intent objects through integrated signal extraction, structured parsing, and intent-driven generation. Its core innovations include formal definitions of “workflow signals” and “structured intents,” along with a lightweight intent abstraction mechanism. Evaluated on a benchmark of 1,000 synthetically generated multi-intent queries, Opus significantly improves the logical coherence, semantic consistency, and semantic similarity of generated workflows—particularly under high-complexity conditions.

0 citationsRead paper

Opus: A Workflow Intention Framework for Complex Workflow Generation

Feb 25, 2025

This paper addresses the challenge of identifying and encoding process objectives in complex business scenarios. To this end, it proposes the Workflow Intention framework—the first to formally define *Workflow Signal* and *Workflow Intention*, and to establish a mathematical representation system comprising signal vectors and intention tensors. Methodologically, it introduces an end-to-end multimodal business artifact encoder that integrates intra-modal attention, cross-modal fusion attention, and a four-stage intention decoding mechanism, augmented by a customized loss function and joint training strategy. The key contributions are: (1) the first generalizable, interpretable, and scalable workflow intention generation system; and (2) empirical validation on real-world business data demonstrating significant improvements in intention recognition accuracy and process generation compliance—thereby enabling automated workflow construction under quality, regulatory, and compliance constraints.

0 citationsRead paper

Recent publications

Latest Papers

Opus: A Quantitative Framework for Workflow Evaluation

Nov 06, 2025

This paper addresses the absence of a unified quality quantification standard for automated workflows. We propose Opus, a novel evaluation framework that jointly models four orthogonal dimensions—correctness (success rate), reliability (structural consistency and information hygiene), efficiency (resource consumption), and value (output gain)—integrating reward mechanisms with normative penalties. Grounded in expected utility theory, structured coupling metrics, observability analysis, and information entropy detection, Opus constructs a probabilistic scoring model. The framework enables automatic workflow scoring, cross-process comparison, and multi-objective optimization, and can be embedded into reinforcement learning loops to support end-to-end workflow discovery and iterative refinement. Experimental results demonstrate that Opus significantly improves both efficiency and reliability of automation systems and accurately identifies Pareto-optimal workflows in complex scenarios.

0 citationsRead paper

Opus: A Prompt Intention Framework for Complex Workflow Generation

Jul 15, 2025

To address the insufficient quality and reliability of LLM-generated workflows for complex, multi-intent user queries, this paper proposes Opus, a prompt-based intent framework that introduces a reproducible and customizable intent-capture layer between natural-language queries and workflow generation. Opus decomposes mixed intents into structured intent objects through integrated signal extraction, structured parsing, and intent-driven generation. Its core innovations include formal definitions of “workflow signals” and “structured intents,” along with a lightweight intent abstraction mechanism. Evaluated on a benchmark of 1,000 synthetically generated multi-intent queries, Opus significantly improves the logical coherence, semantic consistency, and semantic similarity of generated workflows—particularly under high-complexity conditions.

0 citationsRead paper

Opus: A Workflow Intention Framework for Complex Workflow Generation

Feb 25, 2025

This paper addresses the challenge of identifying and encoding process objectives in complex business scenarios. To this end, it proposes the Workflow Intention framework—the first to formally define *Workflow Signal* and *Workflow Intention*, and to establish a mathematical representation system comprising signal vectors and intention tensors. Methodologically, it introduces an end-to-end multimodal business artifact encoder that integrates intra-modal attention, cross-modal fusion attention, and a four-stage intention decoding mechanism, augmented by a customized loss function and joint training strategy. The key contributions are: (1) the first generalizable, interpretable, and scalable workflow intention generation system; and (2) empirical validation on real-world business data demonstrating significant improvements in intention recognition accuracy and process generation compliance—thereby enabling automated workflow construction under quality, regulatory, and compliance constraints.

0 citationsRead paper