Northeastern University Helps Shape the Future of Emergency Preparedness with AI and Simulation Gaming

Here’s how experts from Northeastern University's Institute for Experiential AI transformed a vast array of raw, fragmented data into a structured database and knowledge graph that powered a retrieval augment generation (RAG) AI system for the emergency preparedness non-profit Center for Advanced Preparedness and Threat Response Simulation (CAPTRS).

Led by the institute’s AI + Life Sciences team, the resulting AI system has both streamlined existing operational processes and opened up entirely new opportunities for CAPTRS related to emergency preparedness. The methodologies employed by the AI + Life Sciences team are domain agnostic, allowing them to be applied to similarly unstructured data across diverse fields.

The Challenge

The nonprofit Center for Advanced Preparedness and Threat Response Simulation (CAPTRS) seeks to improve the ability of decision-makers to address threats like pandemics and natural disasters through simulation games. As part of that work, the center creates scenarios designed to mimic actual emergencies, like an infectious disease outbreak. The goal of these scenarios is to challenge decision-makers and force them to reflect on gaps in their own preparedness. Said differently, they want realistic scenarios, but not close replicas of events that have happened in the past. 

Each simulation is the result of a time-intensive process in which experts create unfolding disaster stages, meticulously adding critical details that emergency responders might use to craft a response. Even for seasoned professionals, the sheer volume of historical emergency events can be overwhelming. Consider this: over the past 30 years, the World Health Organization has recorded more than 100,000 infectious disease outbreaks. Manually incorporating learnings from each of these events into new scenarios is infeasible. CAPTRS needed a solution to streamline and scale its scenario-building process—one that could efficiently incorporate vast amounts of unstructured data from past outbreaks while enhancing the depth and accuracy of its simulations.

Partnering with the AI + Life Sciences Team

For this project, CAPTRS leveraged the deep expertise of Northeastern's AI + Life Sciences team, drawing on their specialized knowledge in infectious disease biology and epidemiology. The primary focus was on pathogen threats; however, both the institute and CAPTRS maintain broader research scopes, and the overall approach remained domain-agnostic.

To develop a structured data asset and retrieval augment generation (RAG) model for creating and evaluating pathogen threat simulations, the AI + Life Sciences team, led by Director Sam Scarpino, combines Northeastern’s proven experience in use-inspired research, a world-class co-op program, and a delivery-focused data science team to democratize access to AI technology.

This project benefited from work by Northeastern professors, graduate students from the co-op program, and Institute for Experiential AI data scientists, engineers, and full-stack developers. The multidisciplinary Northeastern experts met regularly with the CAPTRS team to conduct interviews and understand their process for building and evaluating simulations, ensuring their solutions would fit seamlessly into existing workflows.

To build the solution, the AI + Life Sciences team first compiled data from over 100,000 past disease outbreaks and processed it using natural language processing, data mining, and large-scale knowledge curation tools to create a structured database. The database was then integrated into a pre-trained large language model (LLM) using a technique called retrieval augmented generation (RAG), which reduces LLM hallucination by retrieving chunks of text from curated documents – in this case structured data on past outbreaks. RAG has become a state-of-the-art method for improving the reliability and contextual relevance of language model outputs.  

The teams conducted a rigorous testing and evaluation process to assess the model’s ability to generate hyper-realistic scenarios, comparing its outputs to real-world outbreaks.

Delivering Solutions

The database and RAG model built by the Northeastern team offered CAPTRS a way to easily create simulations grounded in truth. The model’s outputs were far more detailed and sophisticated than those created using off-the-shelf LLMs.

The AI + Life Sciences team also built a classification engine that could categorize emergencies and outbreaks based on factors like pathogen type, size, and spread.

CAPTRS uses its simulations in work with leaders in the World Health Organization, public health groups, the Centers for Disease Control and Prevention, and other agencies around the world. The CAPTRS team believes their new tool will expedite the process of creating realistic simulation exercises and help them add richer details to their scenarios.

Conclusion

The CAPTRS project is one of many the experts at the Institute for Experiential AI are building to turn data in the form of PDFs, emails, and other formats into structured, queryable databases that can interface with LLMs to create powerful products, streamline processes, and open up new streams of revenue for partners. Much of that work is done through the institute’s AI Solutions Hub, which works with companies across industries to develop cutting-edge solutions to their data challenges. The institute and CAPTRS continue to expand their partnership and develop new AI solutions.

Learn more about how the Institute for Experiential AI can build custom AI solutions for your business.

‍Learn more about how CAPTRS is harnessing the brightest minds in AI and simulation gaming to build robust defenses against future threats.