Open-source research infrastructure for trauma-informed AI in health and social services.
TIAIL is the research program of Nest and Rise, Inc., a 501(c)(3) nonprofit based in Baltimore, MD. This repository holds the lab's public artifacts as they are released: governance frameworks, evaluation rubrics, model benchmarks, and reference implementations for use by nonprofits, public agencies, and researchers.
The work is public by design.
- Ethical AI principles. The foundational commitments guiding TIAIL's research and engineering decisions.
- Contributing guide. How to participate as a developer, ethicist, clinician, privacy practitioner, or contributor with lived experience of the systems we study.
- Institutional review protocols and deployment criteria for AI in regulated care settings.
- Evaluation rubrics calibrated to clinical and human services settings, including a retraumatization risk scoring protocol.
- Model benchmarks for trauma-informed performance on intake, triage, and case management tasks.
- Open-source toolkits and documentation frameworks for Medicaid, HUD, and grant compliance.
Trauma Informed AI Lab (TIAL) develops governance frameworks and evaluation tools grounded in established standards and adapted for the specific risks of AI systems serving vulnerable populations in healthcare and social services. The work is governance research first: the social mission is delivered through controls, evidence, and auditable design, not sentiment.
NIST AI RMF The Govern, Map, Measure, and Manage functions applied to AI systems serving vulnerable populations. Govern establishes accountability for systems affecting people with limited recourse; Map identifies context-specific harms before deployment; Measure defines fairness and safety metrics for the population served; Manage operationalizes monitoring and intervention thresholds.
ISO 42001 (in progress) AI Management System design adapted for healthcare and social services contexts, where the operating environment includes protected health information, eligibility determinations, and populations who cannot easily contest an automated decision.
Trauma-Informed Design Principles Safety, trustworthiness, peer support, collaboration, empowerment, and cultural sensitivity embedded as governance requirements rather than afterthoughts. Each principle maps to a concrete control: trustworthiness to explainability, empowerment to human-in-the-loop and contestability, safety to harm monitoring and intervention thresholds.
- Bias and fairness in eligibility decisions — AI systems that influence access to housing, healthcare, and social services can produce disparate impact on the populations least able to challenge it. We define fairness metrics and disparate-impact monitoring before deployment, not after harm.
- Data privacy and consent for populations with limited agency — consent frameworks designed for users who may face cognitive, legal, language, or power-asymmetry barriers to meaningful consent.
- Human-in-the-loop controls for high-stakes decisions — required human review at every decision point where the consequence of an error is material to a person's safety, benefits, or care.
- Auditability and explainability for publicly funded AI — systems funded by public dollars must be reconstructable and explainable to a non-technical oversight body, not only to engineers.
- HIPAA — Privacy and Security Rule requirements for AI systems that process, store, or route protected health information, including minimum-necessary access and audit controls for agentic components.
- Federal procurement AI governance — transparency, oversight, and documentation requirements for AI used in publicly funded programs.
- State-level AI regulation — emerging frameworks governing automated decision systems in benefits, healthcare, and consumer contexts.
Feedback and collaboration welcome. jimiige.com | LinkedIn
- Site: https://tiail.org
- Volunteer: https://forms.cloud.microsoft/r/xCedQ68MdC
- Discord: https://discord.com/invite/9vfjCa8Usj
- Newsletter: https://substack.com/@nestandrise
- Donate: https://givebutter.com/InnovationFund
MIT. See LICENSE.