Dual Minds: AI Agents Collaborating for Data-Driven Solutions

Abstract

Organisations in a variety of industries, including retail and healthcare, have depended more and more on data-driven methods in recent years to aid in both strategic and operational decision-making. While traditional data analytics and forecasting methods provide useful information, they often require expert interpretation and cannot provide non-technical users with useful recommendations. Furthermore, existing analytical systems are typically domain-specific and unadaptable when used with heterogeneous datasets. This work describes a multi-domain decision intelligence system that combines forecasting, AI-based advice, and automated exploratory data analysis (EDA) using large language models (LLMs) and retrieval-augmented generation (RAG). The proposed system automatically identifies an incoming dataset\'s domain, infers its schema, and generates understandable data summaries without requiring human configuration. Machine learning-based forecasting is used to identify non-linear trends in domains with rich historical patterns, such retail sales; solid baseline forecasting techniques are utilised for other domains, like healthcare, to guarantee dependability and interpretability .The system integrates domain-specific knowledge obtained from RAG with deterministic EDA insights and forecast summaries to close the gap between analytics and decision-making. An LLM is then given this grounded contextual data to produce trustworthy, comprehensible, and context-aware natural language suggestions .The suggested paradigm minimises hallucinated reactions, facilitates informed decision-making, and adapts well across domains, according to experimental evaluation on retail and healthcare datasets. The findings demonstrate the potential of combining RAG-enabled LLMs with structured data analytics to create reliable and user-focused decision support systems.

Introduction

The text proposes a multi-domain decision intelligence system designed to bridge the gap between raw data analytics and practical decision-making using forecasting models, exploratory data analysis (EDA), and Retrieval-Augmented Generation (RAG)-based large language models (LLMs).

The problem addressed is that modern organizations collect large amounts of data across sectors like retail and healthcare, but existing tools are fragmented. Traditional dashboards provide only descriptive insights, forecasting models provide predictions without actionable guidance, and LLM-based systems offer interaction but may produce hallucinated or unreliable outputs. This creates a gap between analysis and real-world decision support, especially for non-technical users.

To solve this, the study proposes a unified decision intelligence framework that integrates:

• Automated EDA to understand dataset structure, trends, and anomalies
• Forecasting models (Random Forest, LSTM, and baseline methods depending on data richness)
• Domain and schema inference for automatic adaptation across datasets
• RAG-based LLM advisory system that grounds responses in retrieved domain knowledge and computed analytics

The system is designed in a modular architecture with five main stages:

1. Data ingestion and preprocessing
2. Domain and schema detection
3. EDA and forecasting
4. Insight generation in natural language
5. RAG-based decision advisory

It is tested on retail sales and healthcare datasets, showing adaptability across structured, high-volume data (retail) and irregular, low-data environments (healthcare). Retail data supports advanced forecasting due to clear temporal patterns, while healthcare data uses simpler, more reliable models due to variability and limited history.

The literature review highlights that:

• EDA and forecasting are widely used but often isolated
• LLMs improve accessibility but suffer from hallucinations
• RAG improves factual reliability by grounding outputs in external knowledge
• Existing systems rarely integrate analytics, forecasting, and reasoning into one pipeline

Overall, the proposed system enables end-to-end decision support, transforming raw data into:

• automated insights
• reliable forecasts
• and actionable, explanation-based recommendations for non-technical users

Its key contribution is a unified, adaptive, and explainable AI framework that combines deterministic analytics with probabilistic language reasoning to support real-world decision-making across multiple domains.

Conclusion

In order to facilitate data-driven decision-making, this study introduced a multi-domain decision intelligence system that combines forecasting, big language models based on Retrieval-Augmented Generation, and exploratory data analysis. The suggested methodology combines deterministic analytics with grounded language-based reasoning to bridge the gap between data analysis and actionable decision assistance, in contrast to conventional analytical dashboards or stand-alone forecasting models. By identifying domain features and deriving schema information, the system automatically adjusts to diverse datasets, facilitating smooth operation across many domains including retail and healthcare. The system successfully creates interpretable insights, relevant forecasts, and trustworthy natural language recommendations based on both domain-specific knowledge and data-driven summaries, according to experimental evaluation. Retrieval-Augmented Generation greatly decreased hallucinogenic responses and increased the advisory outputs\' credibility. The suggested design guarantees explainability and dependability, which are essential for real-world decision-making situations, by isolating probabilistic language production from analytical processing. The outcomes demonstrate how feasible it is to combine RAG-enabled large language models, automated analytics, and machine learning into a single, user-focused decision support system.

References

[1] J. W. Tukey, “Exploratory Data Analysis,” Addison-Wesley, 1977. [2] L. Breiman, “Random Forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001. [3] ECML-PKDD, “Applied Machine Learning Systems for Data Analytics,” European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2025. [4] P. Harper, “Healthcare Analytics: Methods and Applications,” Springer, 2018. [5] T. Brown et al., “Language Models are Few-Shot Learners,” Advances in Neural Information Processing Systems (NeurIPS), 2020. [6] Z. Ji et al., “Survey of Hallucination in Natural Language Generation,”Association for Computational Linguistics (ACL), 2023. [7] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” Advances in Neural Information Processing Systems (NeurIPS), 2020. [8] Y. Zhang et al., “Theory of Mind for Multi-Agent Collaboration via Large Language Models,” arXiv preprint, 2023. [9] C. Huyen, “AI Engineering: Building Applications with Foundation Models,” O’Reilly Media, 2024. [10] S. Few, “Information Dashboard Design: Displaying Data for At-a-Glance Monitoring,” Analytics Press, 2013. [11] [11] Knaflic, S. N., “Storytelling with Data: A Data Visualization Guide for Business Professionals,” Wiley, 2015. [12] G. Box, G. Jenkins, G. Reinsel, “Time Series Analysis: Forecasting and Control,” Wiley, 2015. [13] Hyndman, R. J., Athanasopoulos, G., “Forecasting: Principles and Practice,” OTexts, 2018. [14] E. Turban et al., “Decision Support and Business Intelligence Systems,” Pearson, 2011. [15] Gao, Y. et al., “Retrieval-Augmented Generation for Large Language Models: A Survey,” arXiv, 2023.

Copyright

Copyright © 2026 Gayatri Sharad Thombare, Dr. Rashmi Thakur. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.