Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Classifier-Augmented Generation for Structured Workflow Prediction

Thomas Gschwind, Shramona Chakraborty, Nitin Gupta, Sameep Mehta

TL;DR

This work addresses translating natural language into executable ETL workflows by introducing Classifier-Augmented Generation (CAG), which decomposes utterances, retrieves candidate stages via a classifier and keyword matching, and guides final predictions with stage-focused prompts. It conducts a comprehensive evaluation across stage prediction, edge layout, and property generation, showing that CAG outperforms single-prompt and agentic baselines while substantially reducing token usage. Key contributions include achieving high stage-prediction accuracy (>97%), around 90% accuracy for property inference, and over 70% structural similarity for full-flow generation, with robust, modular validation suitable for production deployment in IBM DataStage. The results demonstrate a practical, interpretable framework for NL-driven, structured automation tasks, with significant implications for reducing authoring effort and enabling scalable, auditable workflow synthesis.

Abstract

ETL (Extract, Transform, Load) tools such as IBM DataStage allow users to visually assemble complex data workflows, but configuring stages and their properties remains time consuming and requires deep tool knowledge. We propose a system that translates natural language descriptions into executable workflows, automatically predicting both the structure and detailed configuration of the flow. At its core lies a Classifier-Augmented Generation (CAG) approach that combines utterance decomposition with a classifier and stage-specific few-shot prompting to produce accurate stage predictions. These stages are then connected into non-linear workflows using edge prediction, and stage properties are inferred from sub-utterance context. We compare CAG against strong single-prompt and agentic baselines, showing improved accuracy and efficiency, while substantially reducing token usage. Our architecture is modular, interpretable, and capable of end-to-end workflow generation, including robust validation steps. To our knowledge, this is the first system with a detailed evaluation across stage prediction, edge layout, and property generation for natural-language-driven ETL authoring.

Classifier-Augmented Generation for Structured Workflow Prediction

TL;DR

This work addresses translating natural language into executable ETL workflows by introducing Classifier-Augmented Generation (CAG), which decomposes utterances, retrieves candidate stages via a classifier and keyword matching, and guides final predictions with stage-focused prompts. It conducts a comprehensive evaluation across stage prediction, edge layout, and property generation, showing that CAG outperforms single-prompt and agentic baselines while substantially reducing token usage. Key contributions include achieving high stage-prediction accuracy (>97%), around 90% accuracy for property inference, and over 70% structural similarity for full-flow generation, with robust, modular validation suitable for production deployment in IBM DataStage. The results demonstrate a practical, interpretable framework for NL-driven, structured automation tasks, with significant implications for reducing authoring effort and enabling scalable, auditable workflow synthesis.

Abstract

ETL (Extract, Transform, Load) tools such as IBM DataStage allow users to visually assemble complex data workflows, but configuring stages and their properties remains time consuming and requires deep tool knowledge. We propose a system that translates natural language descriptions into executable workflows, automatically predicting both the structure and detailed configuration of the flow. At its core lies a Classifier-Augmented Generation (CAG) approach that combines utterance decomposition with a classifier and stage-specific few-shot prompting to produce accurate stage predictions. These stages are then connected into non-linear workflows using edge prediction, and stage properties are inferred from sub-utterance context. We compare CAG against strong single-prompt and agentic baselines, showing improved accuracy and efficiency, while substantially reducing token usage. Our architecture is modular, interpretable, and capable of end-to-end workflow generation, including robust validation steps. To our knowledge, this is the first system with a detailed evaluation across stage prediction, edge layout, and property generation for natural-language-driven ETL authoring.
Paper Structure (18 sections, 2 figures, 5 tables)

This paper contains 18 sections, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Workflow Generation
  4. Single Prompt Stage Prediction
  5. Agentic Stage Prediction
  6. Classifier Augmented Stage Prediction
  7. Edge Predictions
  8. Property Prediction
  9. Pipeline Summary
  10. Related Work
  11. Limitations
  12. Conclusions
  13. Example Utterances and Flows
  14. Linear Flow
  15. Non-Linear Flows
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Interaction between CAG, edge prediction, and property prediction to generate workflows.
  • Figure 2: Non-Linear Flow Example