Navy - 25.2 SBIR - Generative Artificial Intelligence for Course and Content Creation and Conversion (GenAI4C)

Generative Artificial Intelligence for Course and Content Creation and Conversion (GenAI4C)

Navy SBIR 25.2 - Topic N252-112
Office of Naval Research (ONR)
Pre-release 4/2/25 Opens to accept proposals 4/23/25 Closes 5/21/25 12:00pm ET

[ View Q&A ]

N252-112 TITLE: Generative Artificial Intelligence for Course and Content Creation and Conversion (GenAI4C)

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Human-Machine Interfaces;Trusted AI and Autonomy

OBJECTIVE: Utilize state-of-the-art artificial intelligence (AI) technologies (e.g., Large Language Models) to develop an AI-aided human-in-the-loop instructional systems design agent/coach, support conversion of "legacy" (e.g., PowerPoint and Word) course materials, and assist in generation of content within Marine Corps learning ecosystem (i.e., Moodle) to update/revise or develop new courses.

DESCRIPTION: In January 2023, the Marine Corps released "Training and Education 2030" that begins with the statement "The current [Training & Education] T&E system is not preparing the Marine Corps for the future operating environment." [Ref 1]. Today’s training is built on an industrial-era model where students move through a one-size-fits-all production line, progressing through standardized programs of instruction (POIs) comprised of static slides, student handout documents, written exams, and minimal experiential learning. The same report then goes on to state that "better technology integration in our classrooms and courseware can increase production, produce a more highly trained Marine, reduce their overall time-to-train, and limit the burden on our learning infrastructure."

The recent advancement of AI technologies including Large Language Models (LLMs) and image generators such as Stable Diffusion provide an opportunity to aid Marine Corps transition from industrial to information age learning in three ways: (1) instructional systems design assistance (e.g., develop course outlines, map Knowledge, Skills, and Abilities [KSAs] to tests and quizzes, advise personnel on how to best build instructional exercises in their eLearning environment, etc.) to aid course designers, instructors, and others in revising and bootstrapping course outlines and course content to reduce challenges for personnel (e.g., intimidated by the "blank slate" of creating content from scratch); (2) content generation assistance for creation of multimedia and/or interactive learning aids that includes animations and videos, but also goes beyond the creation of quizzes and flashcards that are possible with today’s LLM technologies to generate branching scenarios, and other more interactive content; and (3) conversion to bring "legacy" content (e.g., text, slides, images, infographics, complex diagrams, etc.) into the modern learning environment.

The overarching goal of this SBIR topic is not to use AI as a standalone replacement for curriculum developers, POI managers, or instructors, but to enable human-AI teams to develop and manage training for the Marine Corps faster and more effectively than current processes. Technology created from this effort is expected to show efficiency gains in content creation and course generation without negatively impacting learning outcomes. Proposed solutions must go beyond the existing commercial and open-source efforts to integrate AI into learning platforms [see Ref 2 as an example]. Proposed solutions must create desired modern, interactive-content AI models that can accept input from a variety of sources and formats and output not just text, but images, movies, and more that must work together seamlessly.

The end state of this effort is to provide a government-owned suite of AI-enabled software capabilities for use by the Marine Corps Training and Education enterprise to more efficiently and effectively create courses and content as part of modern e-learning systems that reflects a cutting-edge, information-age learning enterprise. A good example are Marine Corps maintenance schoolhouses whose courses currently include a long classroom component (e.g., PowerPoint slides and lecture) with some hands-on practical application time (e.g., performing the maintenance task on a physical system). The desire is for the classroom components to be richer with multimedia and interactive components (e.g., branching scenarios allowing students to interactively troubleshoot issues); however, creating these from scratch or converting static text or images is time consuming. Additionally, even converting a Program of Instruction into a new, blank Moodle course is daunting.

The end goal is a tool that helps personnel take their current POI, slides, and documents and creates a new course and populates it with content. Personnel will always be in the loop to verify, modify, and add to AI-created content, but AI technologies that can bootstrap the course and content creation and conversion process would be a significant savings of time and effort.

Due to the potential long review times involved, human subject research is discouraged during Phase I. Phase II plans should include key component technological milestones and plans for at least one operational test and evaluation, to include user testing.

PHASE I: Develop early concepts, wireframes, workflows, and requirements for AI-enabled modern training and education software capabilities to include support for instructional systems design processes (e.g., creating course outlines in eLearning systems based off a Marine Corps POI), content (i.e., multimedia and interactive) creation, and "legacy" (i.e., static PowerPoint and Word) content conversion. Show how the human-AI team will work together to produce training content, since the human designer is an essential part of the development process.

Human factor and human subject testing are critical in follow-on Phases of this topic. Please carefully review the requirements of approval for proposals that include testing of human subject and compliance with Institutional Review Board (IRB) [Refs 3, 4].

PHASE II: Conduct an evaluation with Marines (coordination aided by ONR) of AI-enabled capabilities outlined during the Phase I evaluation to include a usability assessment, process improvement demonstration, and effectiveness evaluations where appropriate. Perform additional demonstrations/experimentation that show system extensibility through plug-and-play of new and updated AI models to improve and expand upon overall system capability. Collect impressions of usability and develop objective metrics of time and effort to create and convert courses and content with a relevant Marine Corps population (i.e., including Marine Corps government and civilians in-the-loop of the course and content creation process). Perform all appropriate engineering tests and reviews, including a critical design review to finalize the system design.

Produce the following deliverables: (1) a working prototype of the system that can interact with existing system specifications; (2) evaluation of system usability and efficiency to convert and create content and courses; (3) a system effectiveness evaluation of system capabilities to provide demonstrable improvements (Human Subjects protocol needs to be approved in Phase II Option if needed for this evaluation).

PHASE III DUAL USE APPLICATIONS: Support the Marine Corps in transitioning the technology for Marine Corps use. Develop the software for evaluation to determine its effectiveness in either a formal Marine Corps schoolhouse or other training setting. As appropriate, focus on broadening capabilities and commercialization plans. Development of affordable, scalable, non-proprietary technologies are needed to accelerate the transition of the Marine Corps to an information age training model.

The commercial sector is developing some of these AI-enabled T&E technologies, but they often do not deal with critical issues regarding non-existent, limited, or low-quality source data, do not address encryption and classification requirements, and often come with prohibitive licensing and usage fees. This technology will have broad application in the commercial sector. Specific examples of businesses in the commercial sector include Moodle (which has a commercial learning services component) and Docebo (an online learning platform for enterprises).

REFERENCES:

"Training and Education 2030." Department of the Navy/United States Marine Corps, January 2023. https://www.marines.mil/Portals/1/Publications/Training%20and%20Education%202030.pdf?ver=G6MJLpoB3_H4JRuo0FwthA%3d%3d
Porritt, Matt. "Announcing Moodle LMS 4.5: Unlocking the power of AI." Moodle Community News, October 7, 2024. https://moodle.org/mod/forum/discuss.php?d=462651
Office of Naval Research. "Human Subject Research Documentation Requirements".

https://www.onr.navy.mil/work-with-us/how-to-apply/compliance-and-protections/research-protections/human-subject-research

Office of Naval Research. "Research Protections". https://www.nre.navy.mil/work-with-us/how-to-apply/compliance-and-protections/research-protections

KEYWORDS: Artificial intelligence; machine learning; AI/ML; content creation; training and education; instructional systems design; large language models

** TOPIC NOTICE **

The Navy Topic above is an "unofficial" copy from the Navy Topics in the DoD 25.2 SBIR BAA. Please see the official DoD Topic website at www.dodsbirsttr.mil/submissions/solicitation-documents/active-solicitations for any updates.

The DoD issued its Navy 25.2 SBIR Topics pre-release on April 2, 2025 which opens to receive proposals on April 23, 2025, and closes May 21, 2025 (12:00pm ET).

Direct Contact with Topic Authors: During the pre-release period (April 2, 2025, through April 22, 2025) proposing firms have an opportunity to directly contact the Technical Point of Contact (TPOC) to ask technical questions about the specific BAA topic. The TPOC contact information is listed in each topic description. Once DoD begins accepting proposals on April 23, 2025 no further direct contact between proposers and topic authors is allowed unless the Topic Author is responding to a question submitted during the Pre-release period.

DoD On-line Q&A System: After the pre-release period, until May 7, 2025, at 12:00 PM ET, proposers may submit written questions through the DoD On-line Topic Q&A at https://www.dodsbirsttr.mil/submissions/login/ by logging in and following instructions. In the Topic Q&A system, the questioner and respondent remain anonymous but all questions and answers are posted for general viewing.

DoD Topics Search Tool: Visit the DoD Topic Search Tool at www.dodsbirsttr.mil/topics-app/ to find topics by keyword across all DoD Components participating in this BAA.

Help: If you have general questions about the DoD SBIR program, please contact the DoD SBIR Help Desk via email at [email protected]

Topic Q & A

4/22/25	Q.	Can you please expand on the expectations of the system's involvement in generating and creating multimedia?
	A.	Please review the SBIR topic call. The objectives are clearly listed.
4/21/25	Q.	What sample datasets are you prepared to provide?
	A.	We hope to provide a high-level overview of potential use cases from Marine Corps schoolhouses during Phase I, but not any specific documents or materials. As such you should plan on having existing samples to use that would relate well to the topic for Phase I activities.
4/21/25	Q.	Can you list the systems that were deemed insufficient for this purpose?
	A.	No
4/21/25	Q.	Would you value the capacity to quickly & automatically scale content for various diverse form factors such as small screen mobile & VR?
	A.	Mobile is a reasonable use case. VR is a far-term Marine Corps capability that we are not focused on at this time.
4/21/25	Q.	The rapid advancement in generative model capabilities significantly outpaces software upgrade cycles, but incorporating new models without rigorous testing for the military professional environment may introduce some mild (but mitigatable) risks of the capacity to generate content of questionable decency. As this solicitation states professional content creation specialists will remain a part of the end state process with this new tool: would you value a proposed system that allows for hot swapping in new models to embrace superior capabilities, alongside this mild risk?
	A.	We are looking for a plug-and-play architecture where different AI systems can serve as the platform since this technology is changing so rapidly. We expect the GenAI4C system to be more extensible than this and be capable of connecting to multiple systems (provided that those systems have an API).
4/21/25	Q.	To what extent should proposed solutions be designed for integration with existing Moodle infrastructure versus standalone functionality in Phase I? Are there specific types of "legacy" course material (e.g., heavily annotated PowerPoint, scanned documents, multi-column PDFs) that are more important for demonstrating conversion capabilities? Is the use of open-source AI models (e.g., Phi, Mistral, LLaMA) preferred or is there flexibility in using hosted APIs for early-stage prototyping in Phase I? What specific multimedia output types (e.g., animations, videos, branching simulations) would be most impactful to demonstrate feasibility within the Phase I timeline? Should proposed systems prioritize modularity for future plug-and-play model updates (as mentioned in Phase II), or is that considered out of scope for Phase I feasibility? Are there baseline instructional design standards or Marine Corps POI formatting expectations that should be followed when generating course structures or mapping KSAs? Is there a preference for edge deployment capability or disconnected operation support in the long-term vision of the tool, or will it remain tethered to cloud-based systems?
	A.	While the Marine Corps currently uses Moodle as their LMS, this could change in the future. Complying with as many standards as possible (e.g., SCORM) is ideal to allow for maximum LMS platform flexibility. PowerPoint, Word, Excel, PDFs, and technical manuals (paper, PDF, and IETM) are likely the most common formats. Each schoolhouse and course develops its own materials, so there is likely a wide variety of complexity and variability in these documents. There will be some standard documents, such as a list of the Program of Instruction, but there will also be a great deal of variability. The specific combination of features is left to the proposer so they can balance technological feasibility with value to the schoolhouses. It is anticipated that this would be the case for Phase II because the end state of the SBIR is that the system must be able to handle CUI (IL4 versus IL5 is not known at this time). Models that run in the commercial cloud (e.g., ChatGPT) will not be allowed because much of the material that Marine Corps Schoolhouses have will be Controlled Unclassified Information (CUI). However, IL4-secured cloud solutions may be developed during the SBIR performance period that would be acceptable. For this project, it would be helpful to start with on-prem models (e.g., a locally-running open-source Llama instance would be acceptable) while understanding how you could easily build out to secure cloud-based methods depending on connectivity. For the final product, DoD-cloud LLMs are likely to be much more powerful and provide better product output, but the roadmap for DoD-cloud LLM integration is much less clear. This is a detail left up to the proposer to determine the mix/capabilities regarding output types. That should be a priority. There will be some standard documents, such as a list of the Program of Instruction, but there will also be a great deal of variability in design standards and course content. For this project, it would be helpful to start with on-prem models while understanding how you could easily build out to cloud based methods depending on connectivity.
4/20/25	Q.	What are the most critical elements of legacy course materials (e.g., PowerPoint slides, Word documents) that the tool should prioritize for conversion to ensure compatibility with Moodle and potential future learning management systems (LMS)? Given the variability in complexity and lack of standardization across schoolhouse materials, are there specific content types (e.g., text-heavy slides, technical diagrams, or embedded quizzes) or metadata (e.g., instructional objectives) that the tool should focus on to maximize efficiency and usability in the converted output? How should the tool balance the generation of new content suggestions (e.g., videos, H5P activities) with the conversion of existing materials to optimize the course creation process? Is there a preferred ratio or workflow for combining converted legacy content with AI-generated enhancements, and how should the tool present these suggestions to human designers for review and integration? What specific usability features or interface design elements are most important for the human-in-the-loop review and editing process to ensure efficient oversight by Marine Corps personnel? Considering the emphasis on easy-to-use tools for human review, are there preferred interface paradigms (e.g., drag-and-drop editors, annotation tools) or cognitive aids (e.g., highlighting suggested changes) that would best support Marines with varying instructional design expertise? What are the minimum requirements for the tool’s plug-and-play architecture to support extensibility with multiple AI models, and are there specific APIs or integration protocols the Marine Corps prefers? Given the expectation for a flexible system that connects to multiple AI systems beyond Moodle’s current limited integrations, are there specific API standards (e.g., REST, GraphQL) or model types (e.g., LLMs, image generators) that the tool should prioritize for compatibility? What are the operational constraints for deploying the tool on on-premises infrastructure, particularly regarding handling Controlled Unclassified Information (CUI) and potential scalability to cloud-based environments? Are there specific hardware limitations (e.g., server specifications), connectivity requirements, or CUI handling protocols (e.g., encryption standards) that the tool must adhere to, and how should it be designed to facilitate a future transition to cloud-based deployment? Can you provide examples as to who specifically are the end users of this technology once it is developed?
	A.	We hope to provide a high-level overview of potential use cases from Marine Corps schoolhouses during Phase I, but not any specific documents or materials. As such you should plan on having existing samples to use that would relate well to the topic for Phase I activities. PowerPoint, Word, Excel, PDFs, and technical manuals (paper, PDF, and IETM) are likely the most common formats. Each schoolhouse and course develops its own materials, so there is likely a wide variety of complexity and variability in these documents. There will be some standard documents, such as a list of the Program of Instruction, but there will also be a great deal of variability. We do not have a preferred ratio - there will be a mix of converting “legacy” content and creating “new” interactive, multimedia content. The expectation is that Phase 1 proposals will show examples of both. There are many possible interaction models that we leave open to the proposer to describe. There are no specific API standards. Model types should be defined by the kinds of capabilities the proposer intends to build into their system. While the Marine Corps currently uses Moodle as their LMS, this could change in the future. Complying with as many standards as possible (e.g., SCORM and xAPI) is ideal to allow for maximum LMS platform flexibility. There are several possible places for this technology to live once deployed, including via MarineNet or as an application on the Marine Corps Virtual Platform. For this project, it would be helpful to start with on-prem models while understanding how you could easily build out to cloud based methods depending on connectivity. For the final product, DoD-cloud LLMs are likely to be much more powerful and provide better product output, but the roadmap for DoD-cloud GenAI integration is much less clear. The system must be able to handle CUI. The intended end user could range from an instructor to a curriculum developer to an enabling learning digital developer. These end users could be active-duty Marines or civilians. ,/ol>
4/10/25	Q.	Would it be of interest to extend the course creation into a personalized llm on a per-student basis to help guide the trainee through a course?
	A.	No, we are specifically focused on using generative AI for course and content creation and conversion.
4/9/25	Q.	1. Given the existing corpus of Marine Corps instructional materials (e.g., legacy PowerPoint, Word documents, technical manuals), could you characterize the semantic complexity, consistency, and variability inherent in these documents, specifically regarding terminology standardization, domain-specific language, and completeness of instructional metadata? 2. Could you describe the Marine Corps’ current Moodle deployment in terms of its infrastructure (server environments, hosting models), integration capability (API availability, supported plug-ins, middleware compatibility), and potential restrictions arising from cybersecurity protocols (DISA STIGs, FIPS standards, data encryption mandates) that might constrain AI-based integrations or automated content uploads? 3. From your perspective, what specific quantitative or qualitative benchmarks (e.g., accuracy thresholds, realism criteria, human-rated trust evaluations) should the generated multimedia training content (animations, branching scenarios, interactive simulations) meet to achieve acceptance within rigorous Marine Corps training and doctrinal validation processes? 4. Considering current instructional design workflows utilized by Marine Corps SMEs and instructional designers, what cognitive or workflow-oriented metrics (e.g., cognitive load theory measures, task completion efficiency, trust calibration metrics) would best reflect an optimal human-in-the-loop balance, ensuring effective yet efficient human oversight of AI-generated instructional content? 5. Can you elaborate on any specific security frameworks, compliance standards, data classification levels, or information assurance protocols (e.g., FedRAMP compliance, DoD Impact Levels, NIST SP 800-171/800-53 requirements, data handling for classified or controlled unclassified information [CUI]) that must be explicitly addressed within our system architecture to enable successful deployment and adoption by the Marine Corps?
	A.	1. Each schoolhouse and course develops its own materials, so there is likely a wide variety of complexity and variability in these documents. 2. There are several possible places for this technology to live, including via MarineNet or as an application on the Marine Corps Virtual Platform. Therefore, there will not be an initial interoperability requirement – there will not be an expectation, for example, of immediate operation within the secure MarineNet environment. 3. The key metric should efficiency gains in course creation (easily and directly measurable by the Gen4AIC system). Learning outcomes are also a key metric, but more difficult to measure. 4. Human-in-the-loop review is absolutely critical. Easy-to-use tools to accomplish the review and editing stages are critical. 5. Systems must be able to handle CUI data. We are working with the Marine Corps Training and Education Command (TECOM) to understand their validation, compliance, and security boundaries especially for emerging technologies and their AI deployment roadmap. For this project, it would be helpful to start with on-perm while understanding how you could easily build out to cloud based methods depending on connectivity.
4/6/25	Q.	I’d appreciate your insights on a few points: How do you prioritize legacy content conversion versus new course creation for maximum impact? Are there preferences or constraints on cloud deployment models (e.g., Azure IL5 vs. on-prem)? Would an Azure OpenAI-enabled Copilot agent for Moodle satisfy the human-in-the-loop design intent? What multimedia and scenario types (videos, branching logic) are most important for GenAI support? What expectations exist for validation, compliance, and security boundaries for M365 CoPilot/Studio in GCC High?
	A.	1. We envision a combination of both. Schoolhouses have course outlines, slides, etc. in Office formats, and converting those into an LMS will alleviate the “tyranny of the blank page.” However, we would expect a GenAI4C system to recommend and create new/revised content based off the conversion (i.e., suggesting a place where a video or an H5P activity would be useful). 2. The Marine Corps Training and Education Command (TECOM) is working on an AI deployment roadmap, though no details can be shared. For this project, it would be helpful to start with on-prem models while understanding how you could easily build out to cloud based methods depending on connectivity. 3. We are looking for a plug-and-play architecture where different AI systems can serve as the platform since this technology is changing so rapidly. Right now, Moodle can only connect to a small number of publicly-available AI models. We expect the GenAI4C system to be more extensible than this and be capable of connecting to multiple systems (provided that those systems have an API). 4. At this stage, we are not specifying which multimedia and scenario types are most important. We are hoping for application across a variety of Marine Corps schoolhouses that teach a variety of topics. 5. We are working with the Marine Corps Training and Education Command (TECOM) to understand their validation, compliance, and security boundaries especially for emerging technologies. For now, any prototype system must be able to handle Controlled Unclassified Information (CUI), but this topic does not expect any compliance or verification work to occur with SBIR funding.
4/4/25	Q.	In terms of learner experience, is there preference for creating SCORM courses that could be put into an LMS like Moodle? Or was this hoping for a different course output format?
	A.	While the Marine Corps currently uses Moodle as their LMS, this could change in the future. Complying with as many standards as possible (e.g., SCORM) is ideal.
4/3/25	Q.	Reference 2 ("Announcing Moodle LMS 4.5: Unlocking the power of AI") shows integration between Moodle and OpenAI. Is there a design goal or requirement related to specific information security? i.e. Are SAAS providers like OpenAI acceptable or would preference be given to an AI system that can be self-hosted within govcloud or on Marine-controlled hardware? Context: As a civilian, I don't know how much of the training material would be subject to data classification rules and would want to account for that in the beginning.
	A.	Systems must be able to handle Controlled Unclassified Information (CUI) data. We are working with the Marine Corps Training and Education Command (TECOM) to understand their validation, compliance, and security boundaries especially for emerging technologies and their AI deployment roadmap. For this project, it would be helpful to start with on-prem AI models while understanding how you could easily build out to cloud based methods depending on connectivity.

[ Return ]