Extracting meaningful information from unstructured medical data represents a critical challenge in contemporary healthcare analytics and research, often demanding significant time, computational resources, and specialized expertise from data scientists, thereby increasing the overall cost and complexity of the process. With the emergence of agentic AI, novel opportunities arise for streamlining and automating complex data processing workflows. We present a comprehensive multi-agent architecture designed to democratize medical data analysis for diverse stakeholders, including data scientists, medical researchers, and healthcare practitioners. Our system enables users to: (i) Gain comprehensive insights into their datasets through automated analysis, (ii) Automatically know more about the data based on the additional support files, and (iii) Develop predictive models without requiring extensive machine learning expertise. The proposed architecture incorporates six specialized agents: (i) ``Type identification agent" that automatically classifies data (structured/unstructured) while implementing privacy-preserving anonymization protocols; (ii) ``Feature identification agent" that extracts features from the data; (iii) ``Feature enrichment agent" that generates contextually relevant keyword vocabularies for each feature present in the dataset based on the user intent; (iv) ``Additional file integration agent" that employs a semantic and keyword based extraction method to incorporate supplementary information about each feature from additional support files (PDF, Excel, CSV); (v) ``Input-output optimization agent" that determines based on the user intent and feature information what would constitute as ideal input and output feature for a machine learning model; and finally (vi) ``Modeling advisory agent" that recommends what model would be best suited for the data. We evaluate our multi-agent system on various types of medical data. For healthcare providers, research institutions, and health-tech companies, this means faster decision-making, lower data processing costs, better compliance with data-handling regulations, and the ability to unlock new revenue streams by turning raw data into actionable insights.
Shimgekar et al. (Mon,) studied this question.