EvolvAI Nexus RFI Response
Title: Community-Centric, Ethical AI for Healthcare: A Response to NIH RFI NOT-OD-25-117
Submitted by: EvolvAI Nexus
Contact: Raj Mallichetty, Chief AI Officer
Email: raj@evolvainexus.org
Date: 06/15/2025
Submitted in response to: NIH Notice NOT-OD-25-117
1. Executive Summary
EvolvAI Nexus is a registered nonprofit organization (501(c)(3)) focused on community-first, ethically aligned AI research and development. We are pleased to submit this response to NIH RFI NOT-OD-25-117, outlining practical, community-centered recommendations for advancing responsible artificial intelligence in biomedical research and healthcare delivery.
Our input reflects our core belief: that equitable, trustworthy AI must be co-developed with communities, grounded in reproducible science, and embedded within transparent governance frameworks. To this end, we highlight the following priority areas:
Foundational Infrastructure: Create an open, federated national AI ecosystem with shared training environments, reproducibility tools, and ethical audit frameworks to empower nonprofits, small innovators, and underserved communities.
Reproducibility & Trust: Enforce standardized reproducibility scorecards and documentation for all NIH-funded AI tools to ensure transparency, cross-site validation, and long-term reliability.
Operational Excellence: Pilot AI solutions to streamline NIH operations — from grant submission to peer review and clinical workflows — with robust metrics for efficiency, accuracy, user trust, and equity.
Validation & Regulatory Collaboration: Develop national testbeds, regulatory sandboxes, and nonprofit-led audit consortia in partnership with FDA, VA, and community health systems to validate clinical AI tools safely and equitably.
Partnerships & Community Stewardship: Foster cross-sector partnerships and governance collaboratives that integrate community voices, clinical expertise, and technological leadership throughout the AI lifecycle.
2. Strategic Architecture: Foundational Themes and Architecture
The foundation of responsible AI in healthcare must rest on accessible, trustworthy, and inclusive infrastructure. EvolvAI Nexus supports NIH’s emphasis on explainability, transparency, auditability, and safety, and urges the agency to lead the development of a national AI infrastructure that is as open and equitable as it is technically sound.
We recommend NIH consider the following key pillars:
A. National AI Infrastructure for Health
Establish a federated, standards-based AI infrastructure that is accessible to researchers, startups, nonprofits, and public health agencies—not just large institutions.
Provide public access to shared model training environments, data governance tools, and privacy-preserving computational resources to enable innovation from low-resource or mission-driven organizations.
NIH should seed the development of “AI Commons” infrastructure hubs, supporting cloud-based environments with pre-approved datasets, reproducibility toolkits, audit logs, and synthetic data generation tools.
B. Modular and Interpretable Architectures
Fund the development and standardization of modular, component-based AI architectures that can be audited, validated, and improved independently across different healthcare settings.
Encourage use of interpretable ML models or hybrid systems that combine statistical robustness with human-interpretable outputs to support clinical adoption.
Promote explainability-by-design, where models are built from inception to support rationale generation, use-case boundaries, and patient-level transparency.
C. Interoperability and FAIR Data Principles
Mandate that funded AI models and tools comply with FAIR principles (Findable, Accessible, Interoperable, Reusable).
Support API-driven frameworks for integrating AI outputs into Electronic Health Records (EHRs) and patient portals, while safeguarding against vendor lock-in.
D. Privacy-Preserving and Community-Sourced Data Strategies
Invest in federated learning, differential privacy, and synthetic data generation methods that allow model training on distributed datasets without exposing sensitive patient information.
Create pathways for community-led data collection and stewardship, ensuring diverse and representative datasets through meaningful partnerships with community health organizations, rural hospitals, and advocacy groups.
E. Continuous Benchmarking and Evaluation
Build infrastructure for live benchmarking of AI models using synthetic test beds, time-sliced real-world data, and clinical simulation environments.
Require documentation of model drift, training history, and performance changes over time in high-stakes biomedical use cases.
At EvolvAI Nexus, we believe equitable access to AI development and validation infrastructure is critical. Without it, community nonprofits and small innovators are excluded from shaping the future of healthcare AI. NIH’s leadership in building this infrastructure will ensure that technical progress is aligned with societal benefit and long-term trust.
3. Operational Excellence
EvolvAI Nexus recognizes that operational excellence is the bedrock of public trust and scientific efficiency. We encourage NIH to harness AI not only in research, but also within its own administrative and service workflows to deliver a more streamlined, equitable, and responsive experience for all stakeholders.
Key opportunities include:
Smarter Grant Submission & Review: Deploy AI-powered pre-submission checks for grant applicants (e.g., budget validation, formatting compliance, plagiarism detection) to reduce administrative errors and reviewer burden. Integrate large language models to assist scientific officers in preliminary triage and categorization of proposals.
Enhanced Peer Review: Pilot AI tools that help reviewers identify methodological flaws, missing citations, or potential conflicts of interest, complementing human expertise rather than replacing it.
Clinical Center Operations: Use predictive AI for scheduling, resource allocation, and patient flow management within NIH clinical centers to minimize wait times and optimize workforce deployment.
Stakeholder Engagement: Employ conversational AI and natural language processing (NLP) to streamline inquiry handling, translate technical jargon for public-facing communications, and support multilingual access to NIH resources.
Recommended evaluation metrics:
For each pilot, NIH should develop an Operational AI Evaluation Dashboard capturing:
Time Savings: Quantitative reduction in processing time for grant applications, review cycles, or administrative tasks.
Accuracy & Consistency: Error rates before and after AI augmentation (e.g., compliance checks, peer review consistency).
User Satisfaction: Stakeholder surveys measuring trust, usability, and perceived fairness of AI-enabled processes.
Cost-Benefit Analysis: Net operational savings versus implementation and maintenance costs.
Equity Impact: Metrics to assess whether AI tools reduce or exacerbate disparities in access, review outcomes, or clinical care timeliness.
Embedding these metrics from the outset will ensure that operational AI pilots align with NIH’s commitment to transparency, accountability, and equitable service for all communities.
4. Facilitating & Validating AI in Healthcare Delivery
Best-practice frameworks, testbeds, and regulatory-science collaborations (e.g., with FDA, VA) to evaluate safety, efficacy, and equity of clinical AI tools
As AI becomes deeply integrated into diagnostic, prognostic, and therapeutic workflows, robust validation and oversight are critical to protect patients and maintain public confidence. EvolvAI Nexus recommends that NIH play a convening role in building a shared national framework for testing and certifying clinical AI systems.
Key recommendations:
National AI Testbeds: Establish funded testbeds where academic researchers, nonprofits, and startups can evaluate clinical AI tools on de-identified, diverse real-world datasets. These should include edge-case scenarios and simulate deployment across varied healthcare settings (urban, rural, underserved communities).
Regulatory-Science Collaborations: Formalize collaborative pathways with FDA’s Digital Health Center of Excellence, the VA, and community hospitals to co-develop regulatory sandboxes. These allow controlled real-world pilots while gathering evidence for safety, effectiveness, and bias monitoring.
Best-Practice Frameworks: Develop NIH-endorsed frameworks covering:
Preclinical Validation: Standardized protocols for algorithm training, test splitting, and external validation.
Continuous Monitoring: Guidelines for post-market surveillance, performance drift detection, and incident reporting.
Equity Auditing: Tools and checklists for developers to audit demographic fairness and unintended biases before and after deployment.
Independent Audit Consortia: Fund independent nonprofit-led consortia to perform third-party audits and public reporting of clinical AI tools, complementing regulatory approvals with community accountability.
By anchoring validation in rigorous science, transparent governance, and cross-agency collaboration, NIH can accelerate the safe, effective, and equitable adoption of clinical AI — strengthening patient care and public trust.
5. Reproducibility and Trustworthiness
For AI to be safely adopted in clinical settings, it must be reproducible, verifiable, and transparent across time, sites, and populations. EvolvAI Nexus supports NIH in championing rigorous practices that make biomedical AI both trustworthy and testable.
We recommend the following strategies:
A. Reproducibility Scorecards
NIH should require every funded AI project to include a Reproducibility Scorecard—a standardized, public-facing assessment tool covering:
Data provenance and versioning (e.g., source, access date, licensing)
Training methods and hyperparameter documentation
Validation strategies (cross-site, multi-institutional, temporal validation)
Model performance under varied demographic or clinical subgroups
Computational environment reproducibility (e.g., containerization, open-source packages)
These scorecards can serve as living quality checklists for reviewers, developers, and regulators, helping NIH set a national benchmark for model reliability. These practices align with emerging international reproducibility and AI ethics standards, ensuring U.S. leadership in trustworthy biomedical AI.
B. Product Documentation Standards
All AI tools developed through NIH funding should include thorough documentation aligned with emerging standards, such as:
Model Cards: Summarizing intended use, performance metrics, ethical considerations, limitations, and caveats.
Data Sheets for Datasets: Providing context about how data was collected, preprocessed, labeled, and governed.
Post-deployment Monitoring Plans: Explaining how real-world performance will be tracked, including model drift detection, incident reporting, and user feedback loops.
These documentation tools should be reviewed not only for completeness, but also for clarity, accessibility, and relevance to clinical users and patient communities.
C. Open and Auditable Tools
To foster confidence, NIH should incentivize the development of open-source or auditable AI systems, allowing third parties to inspect, test, and validate performance and safety. Where full source code is not feasible (e.g., due to privacy constraints), NIH should support code escrow, independent auditing, or red-teaming mechanisms.
6. Partnerships & Ecosystem Building
To ensure ethical, equitable, and impactful deployment of AI in biomedical contexts, NIH must invest in a national ecosystem of collaboration, where innovation, governance, and community trust are cultivated in parallel.
EvolvAI Nexus recommends NIH prioritize the development of an integrated ecosystem with three pillars:
Cross-Sector Partnerships:
Encourage co-led projects between nonprofits, academia, public health agencies, startups, and community health providers.
Provide flexible, seed-level funding to enable nonprofits and smaller organizations to prototype, iterate, and contribute to the national AI infrastructure.
Promote shared governance responsibilities where technology builders, clinical users, and impacted communities shape AI standards and evaluation criteria together.
Distributed Governance Infrastructure:
Governance should be treated as a core function of the AI ecosystem, not a siloed responsibility. NIH should invest in:
Open-source governance toolkits that enable scalable adoption of responsible AI practices.
Regional governance collaboratives hosted by nonprofits, universities, or public health departments to support audit frameworks, community review, and patient voice integration.
Governance mentorships for early-stage AI developers and students to learn how to embed fairness, transparency, and ethical foresight from the outset.
Community-Centric Stewardship:
Equip underserved and underrepresented communities with the tools and resources to participate in AI decision-making—from data ownership to model evaluation.
Include patient advocacy groups and health equity leaders in NIH-funded working groups and advisory boards.
Fund research into sociotechnical dynamics that affect trust, algorithmic harms, and access disparities.
At EvolvAI Nexus, our model reflects this philosophy. We plan to partner with clinicians, student researchers, and community organizations to co-develop responsible AI tools—like our participatory audit framework and symptom-checker prototype (SympAI)—while fostering ethical leadership, reproducibility, and transparency.
We believe NIH has a unique opportunity to create a network of trust and accountability, not just innovation. Embedding governance within partnerships ensures that AI systems are not only effective, but also fair, safe, and worthy of public trust.
7. Closing Remarks
We thank NIH for inviting public responses to shape the future of biomedical AI. EvolvAI Nexus is eager to contribute to a national ecosystem that blends technical innovation with deep ethical responsibility.
We look forward to contributing our expertise and partnering with NIH to translate these recommendations into actionable programs and pilot projects.
We welcome opportunities to:
Join working groups
Collaborate in pilot studies
Provide nonprofit expertise on community-aligned AI tools
Thank you for your leadership and continued investment in advancing responsible healthcare innovation.