LOI Questions and Answers
Questions regarding the Letter of Intent are now closed. The information below reflects responses to questions submitted during the open inquiry period.
| Question | Response |
|---|---|
| Eligibility | |
| Are organizations or teams based outside the United States eligible to apply? | Yes. Participation is open to any organization or institution regardless of location or nationality. Proposed models must be tested on U.S. K–12 student data. |
| Are there requirements about the location, composition, or proportion of the applying team? | No. There are no requirements regarding where prime applicants or team members are based or what proportion of the budget or team must be U.S.-based. |
| What experience and scale requirements must applicants meet? | The lead organization must demonstrate at least one peer-reviewed publication predating May 8, 2026, a track record of contributing significant digital public goods (e.g., a publicly released dataset, open-source model, or evaluation artifacts), and prior deployment or evaluation on real student or user data at meaningful scale. Proof-of-concept or synthetic-data-only work does not satisfy the minimum scale requirement. |
| Are there restrictions on how many proposals an organization may appear on? | No. There are no restrictions on the number of proposals an organization may appear on, whether as a lead applicant or a subaward partner. |
| LOI Process | |
| Can LOI submissions be submitted by email instead of through the form? | All submissions must be made through the Qualtrics form linked in the applicant instructions. Email submissions will not be accepted. |
| Is submitting an LOI a prerequisite for submitting a full RFP proposal? | No. Submitting an LOI is not required. Any eligible organization may submit a full proposal regardless of whether they submitted an LOI. |
| Is there a pass/fail threshold at the LOI stage? | No. The LOI is not a competitive screening process; it is intended to help the funders gauge interest and refine the RFP. |
| Can details change between the LOI and the full RFP submission? | Yes. Applicants may revise or expand upon their LOI responses in a subsequent full proposal, including changes to team composition, partnerships, or technical approach. |
| Team Composition | |
| Is there a limit on the number of PIs or co-PIs? | No. The grant is designed to support multi-organizational, cross-disciplinary teams, and required roles may be distributed across multiple individuals and organizations. |
| Are there institutional title or affiliation requirements for applicants? | No. However, all applicants are subject to the same eligibility requirements regardless of role or affiliation. See “What experience and scale requirements must applicants meet?” for details. |
| Partnerships & Collaborators | |
| Are joint applications from multiple organizations eligible? | Yes, and they are encouraged. Partnerships do not need to be fully formalized at the time of LOI submission. At the RFP stage, named organizations or consultants will be asked to provide short letters confirming their role and willingness to serve. At least one committed or conditionally committed ed-tech partner for Phase 3 integration testing is required at proposal submission. |
| Does the awarded team automatically have access to the Existing Resources projects? | Yes. The awarded team will have access to all Existing Resources datasets, benchmarks, and toolsets regardless of whether those PIs are formally part of the proposal. |
| Scope | |
| Is the grant limited to math tutoring? | Yes. The grant is focused on one-to-one AI math tutoring for U.S. K–12 students, though the model should be designed so that its components are modularized and interoperable for potential extension to other educational interactions. |
| Would a project in which AI supports or coaches a human instructor (rather than serving as the direct tutoring agent for students) be eligible? | No. The AI must serve as the direct tutoring agent for the student, which can occur within a classroom setting alongside a teacher. Projects in which AI supports or coaches a human instructor rather than tutoring students directly are not eligible. |
| Are proposals focused on standalone end-user applications eligible? | No. This opportunity is intended to support foundational infrastructure and shared ecosystem capabilities. Proposals focused solely on point solutions or standalone end-user applications will not be considered responsive. |
| How strictly must the model align with U.S. curriculum frameworks (e.g., Common Core)? | No specific framework is prescribed. The emphasis is on alignment with teachers’ instructional plans and philosophies and on learning science principles applicable to U.S. K-12 math contexts. |
| Technical and Architectural Approach | |
| Is a specific technical approach (e.g., fine-tuning) required? | No. The grant is open to a range of approaches, including but not limited to fine-tuning, reinforcement learning, new model architectures, harnesses, and retrieval-based methods. Applicants are encouraged to describe the tradeoffs of their chosen approach and how they would evaluate the relative contribution of different components. |
| Are harness-based or agentic architectures within scope, and is the primary goal open model weights or open tutoring infrastructure? | Both are within scope. The grant does not prescribe a specific technical path, and anticipated public goods include both model artifacts (weights, training data where permissible, fine-tuning playbooks) and infrastructure (evaluation frameworks, benchmarks, APIs). Applicants may propose any combination of approaches — including harnesses, orchestration layers, retrieval-based methods, fine-tuning, or some combination — and how a project allocates emphasis across those is an architectural choice applicants are welcome to propose and defend. |
| U.S. Student Data Requirement | |
| What does it mean to “test on U.S. K-12 student data”? | Models must be tested on U.S. K-12 student data. Several existing open-access datasets are identified as resources applicants are encouraged to use, including the NTO Million Tutor Moves dataset, the SCALE AI Math Tutoring Benchmark data, and the Polygence/TeachLM corpus. Primary data collection with U.S. students is also expected as part of field testing. |
| If a project is primarily piloted on non-U.S. student data, does that disqualify the applicant? | A pilot conducted exclusively on non-U.S. student data would not meet the eligibility requirements. |
| Data, Privacy, and Compliance | |
| What data governance or privacy requirements apply to entities working with U.S. student data? | All proposals must include a safety and bias mitigation plan. Entities working with U.S. student data are expected to comply with applicable U.S. student data privacy laws, and IRB requirements would be governed by the policies of institutional partners. |
| Are budget lines permitted for U.S. partnership costs (e.g., sub-awards to schools, travel for co-design sessions)? | Yes. Such costs are consistent with the grant’s emphasis on practitioner co-design and real-world classroom testing. |
| Open-Source Requirements and IP | |
| What artifacts are required to be open-sourced, and does that extend to proprietary products built on top of the model? | Grant-funded artifacts (e.g., model weights, training data where permissible, and fine-tuning playbooks) must be publicly released under a permissive license as digital public goods. Proprietary products or platforms built on top of those artifacts are not required to be open-sourced. IP terms governing ownership of grant-funded deliverables will be finalized in the RFP and grant agreement. |
| Project Phasing and Timeline | |
| Are the phases and milestones in the LOI required, or illustrative? | Illustrative. Applicants are invited to suggest revisions to the phasing and sequencing, but proposals should demonstrate a credible path to field testing with real U.S. students within the grant period. |
| Is there flexibility in the project timeline? | Yes. Applicants may propose revised sequencing, but stage-gate reviews are a core feature of the grant design and proposals should include a realistic timeline for reaching field testing within the grant period. |