ARIES: A Scalable Multi-Agent Orchestration Framework for Real-Time Epidemiological Surveillance and Outbreak Monitoring

Aniket Wattamwar, Sampson Akwafuo · 2026

Read ARIES: A Scalable Multi-Agent Orchestration Framework for Real-Time Epidemiological Surveillance and Outbreak Monitoring by Aniket Wattamwar, Sampson Akwafuo (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-01831.txt.

Introduction

Disease surveillance still leans on static, single-disease dashboards while outbreaks move fast and data sits in separate silos. Aniket Wattamwar and Sampson Akwafuo present ARIES, a multi-agent framework that autonomously queries sources such as the World Health Organization, the Centers for Disease Control and Prevention, and peer-reviewed papers.

ARIES is organized as a hierarchy. Language-model agents command specialized sub-agents. Together they extract and synthesize surveillance data. The goal is to flag emerging threats and diverging signals in near real time. The authors argue that a task-specific agent swarm beats general-purpose AI. General models suffer hallucinations in this high-stakes domain. The modular design also extends to new diseases and data sources. That suits next-generation outbreak response.

Catalog record

CREATORAniket Wattamwar, Sampson Akwafuo
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-01831

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

ARIES: A Scalable Multi-Agent Orchestration Framework for Real-Time Epidemiological Surveillance and Outbreak Monitoring Aniket Wattamwar, Sampson Akwafuo Abstract Global health surveillance is currently facing a challenge of Knowledge Gaps. While general-purpose AI has proliferated, it remains fundamentally unsuited for the high-stakes epidemiological domain due to chronic hallucinations and an inability to navigate specialized data silos. This paper introduces ARIES (Agentic Retrieval Intelligence for Epidemiological Surveillance), a specialized, autonomous multi-agent framework designed to move beyond static, disease-specific dashboards toward a dynamic intelligence ecosystem. Built on a hierarchical command structure, ARIES utilizes GPTs to orchestrate a scalable swarm of sub-agents capable of autonomously querying World Health Organization (WHO), Center for Disease Control and Prevention (CDC), and peer-reviewed research papers. By automating the extraction and logical synthesis of surveillance data, ARIES provides a specialized reasoning that identifies emergent threats and signal divergence in near real-time. This modular architecture proves that a task-specific agentic swarm can outperform generic models, offering a robust, extensible for next-generation outbreak response and global health intelligence. I Introduction Over the years, lots of data has been published by the World Health Organization (WHO), Centre for Disease Control and Prevention (CDC) [ 26 ] , and National Library of Medicine (NCBI). This data is available in all types of format like XML, JSON, text and tables. Based on this data, there are numerous dashboards for outbreaks, analysis, and trends. Algorithms like time series forecasting, machine learning algorithms are implemented to predict the nature of the disease spread. WHO published Disease Outbreak News (DONs) [ 27 ] , [ 28 ] frequently with information about the assessment, and advice. These dataset are huge in numbers and it is quite an engineering challenge to curate it for a specific task. A significant effort is required to collect, query, analyse and show the data in the form to make decisions and analyse trends. In real time, it is extremely challenging to provide insightful information by querying multiple sources and gathering text in various formats causing high latency and delaying decision and actions. In recent years, with the development of Large Language Models (LLMs) like GPT by OpenAI [ 19 ] , [ 20 ] it has improved our way of extracting information and data tremendously. LLMs have also been trained on multiple research and literature texts spanning from the most mundane information to all the information available on the internet. LLMs are developed in phase and various LLMs are trained on multiple information sources. The use of LLMs in computational epidemiology and disease outbreak has also increased. Information Extraction of various formats and analyzing it is also another significant engineering challenge [ 14 ] , [ 16 ] . These LLMs are generalized models that have the ability to answer questions based on the user query. They are also prone to hallucinations and wrong factual information or extracting outdated information. Knowledge Gap exists between the LLM model trained and the latest advancement in a particular field. LLMs work better in one specific domain or area. However, these LLMs have an overhead cost associated with it. Regarding its usage in surveillance and outbreak there is a need to have a framework that would allow users to have conversations [ 15 ] in a way that reduces the risk of hallucination, is specified for that task only and retrieves relevant and updated information only. To tackle this challenge, this paper introduced ARIES, which is a multi agent framework for automated disease surveillance. It acts as a Decision Intelligence System with multiple agents performing respective and assigned tasks. The approach is scalable with the current architecture where new data sources emerge then spawning new agents and tools can be incorporated without any redesign. II Related Work Samaei et al. [ 1 ] introduced EpidemIQ which is a multi-agent framework that takes input from users and specific agents to perform tasks like literature review, data visualization, network modeling and expert tasks to create a full manuscript in scientific format. The tool being very powerful is not developed to take into account the recent outbreak news, trends of disease spread. But, this paper introduces that factor of real time outbreak analysis and reasoning using multi agent architecture with state-of-the-art technologies. Large language models are constantly used for multiple purposes from summarization, extraction, generation, and analysis. Kwok et al. [ 2 ] and Patlolla, Padmavath et al. [ 9 ] and others used similar methods to chat with the LLM like ChatGPT to estimate two important epidemiological parameters i.e reproductive number and epidemic size. With natural language generation and clarifying the objectives and refining the responses using the susceptible-exposed-infected-recovered (SEIR) framework a classic disease transmission model can be utilized. This approach although effective limits us to using tools like ChatGPT which still use generalized Language Models not having entire access to the previous data. For complex transmission models, the natural language generation uses the entire context window. The free version limits users to converse for a limited time in the same context. Changing the context, the user loses all the progress done so far. This prevents generating the models or generating inaccurate models. Adopting ARIES makes sure that data is recent since the agents are using tools from the official APIs or data sources. R. Allard et al. [ 3 ] showcases time series on disease surveillance with ARIMA (autoregressive, integrated, moving average) algorithms. The paper concluded with the use of forecasting allowed decision makers to consider variability if not detect outbreaks. However, it does not include intelligence in terms of Generative AI. It focuses more on traditional Machine Learning(ML) algorithms. This is similar to the work by Wattamwar et al [ 7 ] and [ 8 ] , where global outbreaks of viral disease are monitored using a dedicated approach. But with ARIES when a user asks a query it can make predictions similar to forecasting based on the data received from the tools associated with the agents. Before Large Language Models became predominant in most of the domains, machine learning and deep learning played a critical role in finding and detecting patterns. It also meant spending a lot of time in data collection, data cleaning and data manipulation. Syed Ziaur Rahman et al. [ 4 ] and others analyzed multiple algorithms like temporal and spatial prediction models and further with risk prediction models to find the correlation between the disease episodes and the characteristics of it. Deep Neural Networks (DNNs) and long-short term memory (LSTM) models along with Semi-Supervised Learning (SSL) algorithms performed by S. Chae [ 10 ] , Kim, J. [ 11 ] , Abougarair, A. [ 12 ] et al. applied across a wide range of text like social media data, and web articles governed by official bodies show promising results. Consoli, Sergio et al [ 5 ] and others have performed information extraction (IE) on datasets provided by WHO and Disease Outbreak News. The extraction is carried across multiple LLMs like Llama-2-70b-chat, Mistral-7b-openorca, Mpt-30b-chat, Pythia-12b, Gpt-4-32k and more. The paper shows results of the accuracy of these models tested across the ProMed and the WHO DONs datasets. This shows the ability of these models to provide accurate results when models are used in

Read the complete text