Grantee Research Project Results

Final Report: GAITER: Generative AI Training for Emergency Responders

EPA Contract Number: 68HERC25C0016
Title: GAITER: Generative AI Training for Emergency Responders
Investigators: Riley, Jennifer
Small Business: Design Interactive Inc.
EPA Contact: Richards, April
Phase: I
Project Period: December 16, 2024 through June 15, 2025
Project Amount: $100,000
RFA: Small Business Innovation Research (SBIR) - Phase I (2025) RFA Text |  Recipients Lists
Research Category: SBIR - Homeland Security , Small Business Innovation Research (SBIR)

Description:

The GAITER (Generative AI Training for Emergency Responders) project explored a novel approach to scenario-based training by leveraging generative AI to rapidly produce immersive, adaptive content from existing response documentation. The goal of this Phase I SBIR effort was to assess the technical feasibility and instructional value of using AI to automate scenario and media generation aligned with emergency response best practices and Standard Operating Procedures (SOPs).

 

Emergency response training requires solutions that are realistic, flexible, and scalable, but traditional methods such as live exercises and static simulations are often costly, time-consuming, and difficult to tailor to specific roles or evolving risks. GAITER was designed to provide supplemental training to overcome these limitations through a modular architecture that leverages large language models (LLMs) and generative AI to automate scenario creation, adapt to diverse training needs, and reduce reliance on manual content development. This technology aims to provide rapid development of robust, repeatable scenarios with dynamic engagements that effectively train processes, procedures, and cognitive and decision-making skills.

 

Shown in Figure 1, the envisioned system integrates four key components:

• A Setup Portal where instructors input SOPs, incident descriptions, and scenario parameters.
• A Generative AI Editor that creates scenario media and branching decision logic.
• A Simulation Training Module that delivers interactive experiences to trainees.
• An After-Action Review (AAR) Tool for contextual feedback and performance tracking.

Summary/Accomplishments (Outputs/Outcomes):

Key accomplishments of Phase I include:

• Defined and applied evaluation metrics for generative AI components (e.g., video, image, audio, avatars, and speech) based on technical feasibility, instructional value, scalability and cost. This analysis supports tool down-selection and informed integration planning for future phases.
• Development of a scenario authoring testbed that accepts SOPs and incident-specific materials, and uses them to generate structured metadata, event timelines, instructional prompts, and decision points.
• Conducted subject matter expert interviews and walkthroughs to gather domain-specific insights on training needs, validate scenario structure, and assess the relevance and clarity of instructional content.
• Created high-fidelity user interface mockups illustrating instructor and trainee workflows, including scenario configuration, media review, and training module progression.
• Designed a modular system architecture that supports scalable deployment, retrieval-augmented generation, and integration of evolving AI tools, with role-driven training flows and performance tracking.

The scope of Phase I focused on evaluating the core feasibility of GAITER’s approach: automating the creation of immersive, adaptive, and document-aware training content. This phase focused on identifying core requirements, validating critical system components, and assessing whether generative tools could realistically support immersive and contextual training development. The findings from this work demonstrated that GAITER’s approach is both feasible and strategically positioned for further development.

GAITER is positioned to fill a critical gap in the training market. While generative AI tools are rapidly advancing, there remains a need to integrate them into cohesive, learning-science grounded workflows that support scalable, immersive training. GAITER addresses this need through modular components that enable rapid scenario development, adaptive content delivery, and long-term sustainability via integration of evolving AI technologies.

The system is built with a flexible, role-driven architecture. Its SOP-aligned logic and branching instructional framework are extensible to any domain where high-pressure decisions, communication, and coordination are essential. As such, the commercialization pathway includes parallel use cases across federal, state, and commercial markets to support a broader ecosystem of training needs beyond emergency response.

Key Differentiators and Market Positioning:

• Modular architecture for generative AI-driven scenario generation
• Designed for broad, cross-domain applicability in complex training environments
• Supports immersive, adaptive training aligned with real-world SOPs
• Enables role-driven workflows and integrated performance tracking
• Closes a critical training gap by unifying AI tools within a scalable training framework

 

Conclusions:

Phase I demonstrated the technical feasibility and instructional value of using generative AI to automate scenario-based emergency response training. The project validated core components of the GAITER system, including its document-driven authoring pipeline, AI-generated media, and modular infrastructure. Subject matter expert feedback reinforced the realism and relevance of the prototype, highlighting GAITER’s potential as a high-impact training solution. SMEs emphasized (1) the importance of SOP alignment and contextual realism for credibility and adoption, and (2) the system’s value in supporting communication and adaptability under uncertainty. They also suggested enhancements to further feasibility, such as integrating scenario roles to reflect multi-agency coordination and designing prompt scaffolding tools to improve content accuracy.

 

Phase II will build on this foundation by expanding GAITER’s capabilities to support more complex decision-making and broad training outcomes. Enhancing branching logic to reflect realistic consequences will enable more nuanced scenarios. The training scope will grow to include interpersonal and coordination skills, allowing GAITER to meet a wider range of training needs. Full system integration will connect authoring, delivery, and after-action review into a seamless pipeline for real-time tracking, feedback, and longitudinal learning. The next phase will deepen GAITER’s impact and enable scalable deployment across EPA and other mission-critical domains.