OPA – Observe, Preprocess, and Act: A Multi-Agent Framework for Data Preprocessing and Predictive Intelligence

ACL ARR 2026 January Submission3713 Authors

Keywords: LLM-Agents, AutoML, Times Series, Forecasting, Multi-Agents

Abstract: The growing availability of temporal data across scientific, industrial, and governmental domains has increased the demand for scalable and reliable time series forecasting solutions. Despite advances in deep learning models and automated machine learning (AutoML) frameworks, building an effective forecasting pipeline remains a complex and decision-intensive process, often requiring expert knowledge and extensive manual intervention. To address these challenges, we propose Observe, Preprocess, and Act (OPA), a multi-agent forecasting framework that autonomously constructs end-to-end predictive pipelines for tabular time series data. OPA leverages recent advances in large language models and agentic artificial intelligence to coordinate specialized agents responsible for data inspection, preprocessing, model configuration, training, and evaluation, all supervised by a central orchestration agent. Users can interact with the system through natural-language prompts to specify objectives and constraints, or fully delegate decision-making to the agents when domain knowledge is unavailable. Experimental results show that OPA achieves forecasting performance comparable to standard AutoML pipelines while significantly improving transparency and interpretability through comprehensive, human-readable reports that explicitly document each modeling decision. These results demonstrate the potential of LLM-driven multi-agent systems to enhance the accessibility, explainability, and usability of AutoML for time series forecasting.

Paper Type: Long

Research Area: AI/LLM Agents

Research Area Keywords: LLM/AI agents, Interpretability, retrieval-augmented generation, applications, reasoning

Contribution Types: Model analysis & interpretability, Data analysis

Languages Studied: English

Submission Number: 3713