Research
from the Opus team.
Technical papers on workflow generation, intention frameworks, quantitative evaluation, and proof-carrying workflow graphs. Each paper is hosted on arXiv or Zenodo.
Opus: Proof-Carrying Streamed Workflow Graphs
Incremental Workflow Graph construction framed as a constrained graph-extension problem. Proof-carrying streams attach a certificate to every admitted addition, witnessing that at least one acceptable completion remains — and viability residuals give the coarsest exact abstraction of that guarantee.
arXiv:2511.04220 · cs.AIOpus: A Quantitative Framework for Workflow Evaluation
A probabilistic-normative formulation for quantifying Workflow quality and efficiency. It integrates correctness, reliability, and cost into a coherent mathematical model that enables direct comparison, scoring, and optimisation of Workflows — and supports automated assessment, ranking, and Reinforcement Learning loops.
arXiv:2507.11288 · cs.AIOpus: A Prompt Intention Framework for Complex Workflow Generation
An intermediate Intention Capture layer between user queries and Workflow Generation. Extracts Workflow Signals from user queries, interprets them into structured Intention objects, and uses them to drive Workflow Generation — yielding consistent improvements in semantic similarity over a 1,000-pair benchmark.
arXiv:2502.19532 · cs.AIOpus: A Workflow Intention Framework for Complex Workflow Generation
A framework for identifying and encoding process objectives in complex business environments. Workflow Intention is the alignment of Input, Process, and Output elements interpreted from Workflow Signal inside Business Artefacts — formalized as a tensor and resolved by an attention-based multimodal generative system.
arXiv:2412.00573 · cs.AIOpus: A Large Work Model for Complex Workflow Generation
The foundational paper on Opus. Introduces a two-phase framework — Workflow Generation via a Large Work Model informed by a Work Knowledge Graph, then Workflow Optimisation via path optimisation on Workflow Graphs. Opus Alpha 1 outperforms state-of-the-art LLMs by 38% / 29% on a Medical Coding use case.