Vasl's research program is in its early stages — which is the honest description of where we are. We have one active IRB study, one active data collaboration with Mayo Clinic Platform, and a clinical advisory relationship with Johns Hopkins. Peer-reviewed publication is in progress. That's what we know. It's enough to build on.
Both are operational — not announced, not aspirational. The University of Maryland IRB study is in active data collection. The Mayo Clinic Platform_Accelerate collaboration is in its first year of a two-year arc. These are the two research foundations that underpin every VLAP accuracy claim on this site.
An IRB-approved study validating CulturalBERT-VLAP's signal detection accuracy against clinician-adjudicated ground truth assessments. The study uses production deployment data from live Vasl cohorts — real members, real language, real clinical annotations by licensed clinicians with community competency training. This is the primary independent validation of VLAP's culturally specific signal detection capabilities. Results will be published in a peer-reviewed journal upon study completion.
Vasl Health is an active participant in the Mayo Clinic Platform_Accelerate program. The collaboration provides access to de-identified behavioral health data infrastructure for VLAP model development and validation. The two-year arc is structured specifically to support CulturalBERT-VLAP's NLP training, validation, and LEP (Limited English Proficiency) population baseline work — with Year 1 focused on model development and Year 2 on prospective validation.
The research program is grounded in three active institutional relationships — each with a specific, documented role. These are not advisory relationships in name only. Each involves ongoing operational work, clinical oversight, or data access that directly informs VLAP's development and validation.
Access to de-identified behavioral health data infrastructure for CulturalBERT-VLAP model development and validation. Two-year structured collaboration — Year 1 model development, Year 2 prospective validation. Provides the data scale necessary for VLAP's NLP training on longitudinal behavioral health signals at clinical population level.
Independent academic validation of VLAP signal detection accuracy through an IRB-approved study comparing CulturalBERT-VLAP output against clinician-adjudicated ground truth. The study represents the first formal independent validation of VLAP's culturally specific signal detection capabilities. Results will be published in peer-reviewed literature upon completion.
Panagis Galiatsatos, MD, MHS — Assistant Professor of Medicine at Johns Hopkins University School of Medicine — serves as Vasl Health's Senior Medical Advisor. Dr. Galiatsatos provides clinical oversight on VLAP's signal detection methodology, accuracy validation approach, non-diagnostic output framing, and the clinical governance of the platform's care coordination model. Active involvement — not nominal affiliation.
Four principles that govern every research decision Vasl makes — from the design of the IRB study to the way we report preliminary findings. These aren't aspirational values. They're constraints that shape the work.
Every outcome claim on this site is drawn from actual pilot data or active IRB research, not projected from internal estimates or aspirational targets. Where findings are preliminary — specifically VLAP signal accuracy — that is stated explicitly, with the caveat that results are pending peer-reviewed publication. We don't report IRB study findings before the study is complete.
The VLAP training corpus was built in partnership with the communities whose language it reads — not assembled from web-scraped data and checked with communities afterward. Vocabulary candidates were community-sourced. Annotation protocols were reviewed by clinicians with documented community competency. Research priorities are informed by community needs. This is the structural difference between studying a community and studying with it.
False positive and false negative rates are disaggregated by demographic subgroup throughout model development — not checked after a model version is deployed. A model version that meets aggregate accuracy targets but fails subgroup parity thresholds across BIPOC, LGBTQ+, or first-generation subgroups is not deployed. This standard is applied to the IRB validation study as a primary research outcome, not a secondary analysis.
All research involving human subjects data — including the University of Maryland validation study and any future studies using Vasl deployment data — operates under IRB approval with informed consent, data privacy protection, and ongoing compliance monitoring. Institutional oversight is a minimum standard, not a credential for marketing purposes. The full IRB study protocol is available to qualified institutional evaluators under NDA.
Vasl Health does not currently have peer-reviewed publications. The IRB study with the University of Maryland is the primary vehicle for the first peer-reviewed publication. Here is what is in progress and when it is expected.
The primary planned publication is the peer-reviewed results of the active University of Maryland IRB study — validating CulturalBERT-VLAP's signal detection accuracy against clinician-adjudicated ground truth in deployed pilot cohorts. The study uses production data from live Vasl deployments across BIPOC, LGBTQ+, and first-generation youth populations. Results will be submitted to a peer-reviewed journal upon study completion.
Study protocol and preliminary design documentation are available to institutional evaluators under NDA. Contact research@vaslhealth.com to request access. We will update this page when the study is submitted for publication and when it is published.
We are actively interested in research partnerships with academic institutions, health systems, and community organizations working on behavioral health equity, culturally-responsive AI, and clinical NLP. We bring the platform infrastructure, production deployment data, and community relationships. We're looking for partners who bring research rigor, clinical expertise, and institutional oversight.
Production deployment data from live Vasl cohorts — BIPOC, LGBTQ+, and first-generation youth populations. The CulturalBERT-VLAP model and its 42-signal taxonomy. Community relationships built through years of co-design. The Mayo Clinic Platform data infrastructure for behavioral health signal data at scale.
IRB-approved research protocols. Clinical expertise in behavioral health, NLP for mental health, or health equity research. Academic partners with experience in community-based participatory research. Peer-reviewed publication capacity and journal relationships in behavioral health, health equity, or clinical AI.
37-page VLAP technical specification. University of Maryland IRB study protocol. Signal taxonomy with full annotation guidelines. Bias monitoring methodology and demographic disaggregation approach. Pilot cohort outcome data (aggregate, de-identified). Mayo Clinic Platform_Accelerate collaboration brief.