Standard NLP models are trained on majority-White internet text. They were never designed to read how Black and Latino, LGBTQ+, and first-generation youth communicate distress — and they consistently fail to. VLAP was built specifically for those communities. Not adapted. Built.
On high-distress signal detection in active IRB validation with a university research partner. Specificity and predictive-value results will be published with the completed study — because half a statistic isn’t transparency.
General-purpose NLP models — including large language models — are trained on internet text that skews heavily toward majority-White, educated, English-speaking populations. The result is a systematic blind spot: the specific ways Black and Latino, LGBTQ+, and first-generation youth signal emotional distress are consistently misread, deprioritized, or missed entirely.
This isn't a model failure in the traditional sense. These models perform well on the populations they were trained for. The failure is in deploying them, uncorrected, for populations they were never trained on — and expecting accurate clinical signal detection to follow.
The gap isn't theoretical. In clinical practice, it means a youth saying "lowkey been struggling fr" is read as casual. A pre-disclosure minimization pattern ("it's not that deep but...") reduces the model's confidence instead of increasing it. Community-developed coded language — terms built specifically to avoid content filters — registers as noise. Standard models consistently miss these signals in the communities that most need them caught.
VLAP analyzes the language members choose to share on the platform — check-ins, peer-group messages, coach conversations. It looks for 47 clinical signals across 6 categories: patterns like pre-disclosure minimization, withdrawal, and escalating distress, expressed the way young people actually express them.
Every detection is sorted into four distinct layers: cultural markers (expression that’s normal in a community’s context), linguistic patterns, behavioral indicators, and clinical relevance. Cultural expression is never treated as a symptom. That separation is the whole point.
It never diagnoses. It never labels or scores a young person. It never sends alerts to police, principals, or parents. It never shows individual conversations to schools or funders. Every escalation goes to one place only: a licensed clinician, who reviews the full context before anything happens.
Members see their own experience. Coaches see the context they need to reach out well. Licensed clinicians see signal context before sessions. Organizations see only aggregate, de-identified trends — never an individual young person.
Our signal accuracy and member outcomes are being independently validated through an IRB-approved study with a university research partner. We publish our measurement definitions, and when the study concludes, we’ll publish what it finds — including accuracy across demographic groups.
Written for a parent as much as a clinician — see the plain-language version for families.
VLAP is a fine-tuned BERT-family transformer model, trained on a purpose-built corpus of culturally specific mental health language. The BERT-family architecture was selected for its bidirectional context processing — essential for reading the layered meaning in code-switching, minimization patterns, and culturally framed expressions where individual words carry different weight depending on surrounding context. The model was then extended with a culturally specific vocabulary and fine-tuned against a clinically annotated training corpus.
The bidirectional encoding layer processes the full context window simultaneously in both directions — unlike unidirectional models that read left-to-right. This is architecturally necessary for cultural signal detection: the meaning of "lowkey" depends entirely on what follows it, and the clinical significance of a minimization hedge ("it's not that deep but") only becomes clear in context of what comes after the conjunction.
Standard BERT-family vocabulary was extended with 2,400+ AAVE terms, youth vernacular expressions, code-switching patterns, and community-developed coded language — including terms built specifically to circumvent content filters (e.g., "unaliving"). This extension was compiled through community engagement with BIPOC and LGBTQ+ youth populations and validated by licensed clinicians with relevant community competency. Without this extension, approximately 23% of the training corpus would be processed as unknown tokens by the base model.
The extended BERT-family model was fine-tuned on annotated language samples drawn from the communities VLAP serves — not web-scraped text, but specifically collected and clinically annotated mental health communication. Each sample was labeled by licensed clinicians with community competency training against the 47-signal taxonomy across six clinical categories. The fine-tuning process was conducted in multiple phases with inter-annotator agreement validation.
The model output layer generates dimensional signal profiles across the six-category taxonomy — not diagnostic outputs, not severity scores, and not clinical recommendations. The output is an interpretive context package: which signal patterns are present, their dimensional classification, and cultural interpretation notes that help clinicians understand what the patterns typically indicate in the communities that produce them. Every output includes a non-diagnostic disclaimer and is designed to support clinical judgment, not substitute for it.
VLAP processes member language in-memory. Verbatim input text is not stored after signal profile generation. The retained output is a dimensional signal profile — not a transcript, not a quote, not a record of the member's exact language. This architectural constraint is a HIPAA technical safeguard, not a configurable setting. It cannot be disabled by organizational administrators or overridden by clinical staff.
The training data for VLAP was not web-scraped, crowd-sourced generically, or assembled from existing open datasets. It was specifically collected from and with the communities the model serves — and annotated by licensed clinicians who have worked within those communities.
Language samples were collected in partnership with community organizations serving Black, Latino, LGBTQ+, first-generation, and youth in urban and rural communities — not scraped from public platforms. Consent protocols, anonymization procedures, and community review were built into the collection process.
All training samples were annotated by licensed clinicians with documented community competency training in the relevant population cohorts. Inter-annotator agreement was validated across the annotation cohort before samples were included in training data.
The 2,400+ vocabulary extension was compiled through structured engagement with youth from the target communities — not through generic web scraping. Vocabulary candidates were reviewed for clinical accuracy by the annotation cohort before inclusion.
False positive and false negative rates were disaggregated by demographic subgroup throughout training — not as a post-hoc audit but as an operational gate. Signal detection performance must meet minimum parity thresholds across Black, Latino, LGBTQ+, first-generation, and limited English proficiency subgroups for a model version to be deployed.
The VLAP Distress Signal Taxonomy (v1.0) is the clinical reference document behind VLAP — it defines the 47 discrete linguistic, behavioral, and contextual signals VLAP is trained to detect, organized into six clinically grounded categories. It serves as the model specification, the annotation codebook for IRB study data labeling, and the signal-level evidence base for VLAP’s NIMH SBIR Phase I application. The taxonomy is not static — it is updated as language evolves and as clinical review proceeds.
VLAP outputs are clinical decision support signals. No signal in this taxonomy, individually or in combination, triggers an automated clinical action. All VLAP outputs are reviewed by a licensed clinician, certified coach, or trained care coordinator before any clinical response. VLAP flags; humans act.
The Beck Hopelessness Scale and Columbia Suicide Severity Rating Scale both identify hopelessness as the strongest independent predictor of suicidal ideation. Generic NLP models perform adequately on Standard American English expressions of hopelessness (HOP-01). VLAP’s differentiation is in detecting culturally coded and vernacular expressions of futility (HOP-02 through HOP-09) that generic models systematically misclassify as frustration or hyperbole.
Standard SAE expression of hopelessness. Well-detected by generic NLP models. Included for completeness and as inter-rater reliability baseline in validation study.
Ambiguous temporal framing. Generic models frequently classify as planning or travel intent rather than ideation. Clinical significance depends on context — requires co-occurring signal for High risk classification.
CRITICAL — highest false negative rate in generic NLP models. The intensifier 'fr' (for real) and double negation construction 'no more' are systematic misclassification sources. Generic models return 'frustration' or 'venting'. VLAP training data includes 340+ AAVE futility constructions. This single signal accounts for the largest share of VLAP's sensitivity advantage over baseline.
In isolation: Low risk, common expression. In combination with ISO or SHA signals: Moderate-High. 'Tired of waking up' is a clinically significant formulation — the exhaustion is specifically about existing. Requires attention to the object of the exhaustion, not just the emotion.
Shift from future-orientation to present-only or past-only temporal framing. Requires longitudinal signal comparison — most meaningful when contrasted against previous session content. Flags for V2 cross-session modeling. In single-session context: Low-Moderate.
Speaking about oneself in the past tense in a living context. Dissociative marker documented in pre-suicidal ideation literature. Distinct from healthy reflection — the distancing is from a valued self-identity. Particularly significant in combination with PFE signals.
Classic pre-suicidal cognition — perceived burdensomeness is one of the two primary components of Joiner's Interpersonal Theory of Suicide (alongside thwarted belongingness). High independent risk. Documented as having the highest correlation with suicide attempt in the clinical literature for this population.
Specifically references the cultural narrative of LGBTQIA+ suffering — 'people like me' and 'people like us' are key referents. Generic models do not detect the referent. For LGBTQIA+ youth, this expression carries the weight of observed community trauma, not just personal pessimism. Requires LGBTQIA+ identity context flag from member profile.
Specific to first-generation and immigrant youth — the weight of intergenerational sacrifice creates a distinct form of hopelessness in which the member perceives themselves as failing a familial obligation. Not detectable by generic models without cultural context. Requires first-gen or immigrant identity flag from member profile.
The SHA category captures both explicit and coded expressions of suicidal ideation. Generic models perform well on explicit ideation (SHA-01) and poorly on all other signals. SHA-02 through SHA-08 represent the core of VLAP’s clinical differentiation — these are the expressions young people actually use, developed specifically to evade platform content moderation and the perceived stigma of explicit disclosure.
Well-detected by all NLP models. Included as validation baseline. VLAP does not differentiate meaningfully from generic models on this signal. The 47-signal taxonomy is justified by performance on SHA-02 through SHA-08 and the other five categories, not SHA-01.
CRITICAL — youth-specific euphemism developed explicitly to evade content moderation on TikTok, Instagram, and other platforms. First documented in widespread use approximately 2020. Most generic NLP models trained before 2022 do not have 'unaliving' in clinical context in training data. VLAP's training data includes 180+ documented uses with clinical annotation. This is one of the clearest demonstrations of generic model failure and VLAP's advantage.
Common expression chronically underweighted in generic models. The sleep framing is literal — the desire is for non-existence, not rest. 'I just want it to stop' requires object identification — what 'it' refers to determines clinical significance. Requires co-occurring signal for High classification unless the expression is highly specific.
Passive ideation — the member is not describing an action but imagining a state of non-existence. Generic models frequently classify as daydreaming or escapism. In clinical literature, passive ideation is a documented precursor to active ideation. Escalate to Moderate when combined with any HOP signal.
CRITICAL — the minimization wrapper ('just wondering', 'not that I'd do it') is the disclosure suppression mechanism, not evidence of low risk. Generic models weight the minimization wrapper and reduce risk classification. VLAP is trained to detect the method inquiry independent of the wrapper. This signal should always trigger human review regardless of stated intent.
Behavioral markers described in text — the member is describing actions consistent with pre-suicidal behavior. Often expressed with a sense of resolution or relief, which is a clinically significant affect shift. Requires careful contextual interpretation — these expressions can be innocent. Combination with HOP-01 through HOP-08 significantly elevates risk.
References to specific dates or events with framing that implies a terminus — a point after which the member does not expect or want to continue. Distinct from goal-setting. The inversion is key: the date is a deadline, not a milestone.
Community-specific coded language for death or suicide. 'Catching a fade' is documented in AAVE; 'checking out' and 'ending the run' are youth-vernacular constructions. These do not appear in generic NLP training data in clinical context. VLAP's training data includes 95+ community-specific death euphemisms with clinical annotation.
CCM signals are the category most specific to Vasl’s target population and most absent from generic NLP training data. They represent culturally conditioned patterns of downplaying distress before disclosing it — a documented phenomenon in Black, Latino, and LGBTQIA+ communities where mental health help-seeking carries stigma. Generic models classify CCM signals as low risk. VLAP treats them as pre-disclosure markers that warrant gentle engagement prompts and session logging.
The minimization opener predicts significant disclosure in the sentence that follows. Generic models weight the minimizer and under-score the disclosure. VLAP is trained to detect the opener as a pre-disclosure flag and analyze the following content at elevated weight. The minimizer is a social safety mechanism — the member is testing for a non-judgmental response before fully disclosing.
AAVE-specific minimization — 'lowkey', 'kinda', and 'ion' (I don't) are qualifiers that reduce apparent severity without reducing actual severity. 'It's whatever though' is a disclosure-withdrawal marker — the member raised something and then minimized. VLAP detects the content of the disclosure, not the qualifier. High false negative rate in generic models.
Pre-disclosure social negotiation — the member is seeking explicit permission and a non-judgmental response before disclosing. Particularly significant in youth who have previously disclosed to an adult and experienced a negative response. VLAP flags this for an immediate warm, non-judgmental coach response.
CRITICAL — humor framing is the most common disclosure suppression mechanism in youth aged 13-22. The joke is the test: if the listener laughs or deflects, the disclosure is abandoned. VLAP is trained to detect the clinical content independent of the humor wrapper. 'Not to be unalive about it' is a specific formulation that combines SHA-02 with CCM-04 — extremely high combined risk.
Distance-creating framing used to test response before self-disclosure — a documented pre-disclosure mechanism in adolescents. VLAP flags all 'asking for a friend' formulations for gentle direct engagement rather than a literal, accusatory response.
The member raised a distress topic and then withdrew from it. This is a failed disclosure — the member attempted to disclose and retreated, likely due to fear of the response. VLAP flags topic drops for coach follow-up: acknowledging the withdrawal and creating a safe reopening.
Internalized stigma — the member has accepted the message that their distress is not legitimate or warranted. Particularly documented in Black, Latino, and LGBTQIA+ youth. The comparison to 'other people' with more severe situations is a disclosure suppression mechanism. VLAP flags for validation-focused coach response.
Strength-performance in communities where emotional vulnerability is stigmatized as weakness — documented in Black male youth in particular, and in communities with strong 'we don't ask for help' cultural norms. VLAP flags for gentle, direct engagement that validates strength while creating space for vulnerability.
Social isolation is a documented independent risk factor for suicidal ideation and is one of the two primary components of Joiner’s Interpersonal Theory of Suicide (thwarted belongingness). ISO signals are most clinically significant in combination with HOP or SHA signals, and as longitudinal markers of increasing withdrawal across sessions.
Explicit social withdrawal. Well-detected by generic models in standard formulation. VLAP adds detection of AAVE and youth-vernacular variants. Moderate risk in isolation; elevates significantly in combination with HOP or SHA signals.
Active severance of social support — distinct from passive withdrawal. The member is taking action to remove supportive contacts. High combined risk when co-occurring with any SHA signal.
Perceived burdensomeness expressed through social withdrawal rationale. Distinct from HOP-07 (self-directed) — ISO-03 is expressed as a reason for social withdrawal rather than a statement about the member's worth. Combination elevates to High.
Documented psychological stressor specific to Black, Latino, and LGBTQIA+ youth — the exhaustion of constantly adapting one's language, presentation, and identity to majority contexts. In VLAP's context, code-switching exhaustion is an isolation marker.
Feeling rejected or invisible within one's own cultural community — a specific form of isolation particularly painful for youth who experience intersectional marginalization. Generic models do not detect the cultural referent.
Family rejection is the single strongest risk factor for LGBTQIA+ youth suicidal ideation — rejected LGBTQIA+ youth are 8.4 times more likely to attempt suicide than accepted peers (Ryan et al., 2009). VLAP flags any expression of family rejection or housing instability related to LGBTQIA+ identity as High risk immediately.
Withdrawal from digital social spaces — for youth, digital social networks are primary social infrastructure. Low risk in isolation; Moderate in combination with ISO-01 or any HOP signal.
TRM signals capture social determinants of mental health that are disproportionately present in Vasl’s target population. Most TRM signals are not independently actionable clinical risk indicators — they are contextual elevators that increase the clinical significance of co-occurring HOP, SHA, or ISO signals. They are also the signals most normalized in community language and therefore most frequently missed by clinicians and tools not calibrated for this population.
Repeated exposure to community violence is a documented trauma and acute stressor for urban youth. For many members, violence is normalized in language. VLAP flags for trauma-informed engagement. Cumulative exposure tracked longitudinally.
Housing instability is a strong predictor of youth mental health crisis, often expressed matter-of-factly in communities where it is common. VLAP flags for social determinants of health documentation and connection to housing support resources.
Immigration anxiety is an acute and chronic stressor for immigrant youth and families. Not independently a clinical risk indicator but a significant context marker that elevates other signals.
Food insecurity is a significant stressor frequently normalized in community language. VLAP flags for connection to food resources and documentation as a social determinant. Low independent clinical risk; contextual elevator.
Police contact and legal system involvement are acute and chronic stressors for Black and Latino youth disproportionately. The fear of police contact is independently traumatic regardless of actual contact.
School push-out is a documented trauma, disproportionate for Black students and students with disabilities, and part of the school-to-prison pipeline stressor. VLAP flags for engagement and connection to educational support.
Cumulative community loss — the accumulation of grief across multiple losses within a community. Particularly significant for Black and Latino youth in communities with high rates of gun violence.
Parentified youth — children assuming caregiver roles — experience heightened stress, accelerated loss of childhood, and reduced access to their own emotional needs. VLAP flags for acknowledgment of the burden and connection to youth support resources.
PFE signals are not distress signals in isolation — they indicate that a previously present protective factor has been removed, which elevates baseline risk. They are most clinically significant as longitudinal markers (V2 cross-session modeling) and in combination with HOP or SHA signals. The protective factors most relevant to Vasl’s population — faith community, mentors, peer networks, cultural identity, and aspirational goals — are different from those in the clinical literature developed for White, middle-class youth populations.
In communities where religious or spiritual practice is a primary coping resource and social support system, loss of faith represents the loss of a primary protective factor AND a primary social network simultaneously. Generic models do not identify this as a clinical signal.
Loss of a trusted adult is a documented protective factor erosion. For youth with limited trusted adult relationships, losing one trusted adult removes a disproportionate share of the member's social safety net. VLAP flags for immediate coach relationship reinforcement.
Loss of peer social support. Distinct from ISO signals (which describe withdrawal) — PFE-03 describes an external dissolution of a previously present support network. Elevates significantly with HOP signals.
Loss of the role or achievement identity that anchored the member's sense of self and social belonging. In combination with HOP signals, this pattern correlates with acute risk.
Aspirational identity erosion. Particularly significant for first-generation youth for whom educational aspirations carry the weight of family sacrifice. A longitudinal hopelessness marker.
Cultural identity is a documented protective factor for BIPOC youth. Strong cultural identity correlates with lower rates of depression and suicidal ideation in these populations. Particularly relevant in first-generation, immigrant, and mixed-heritage youth.
Loss of daily structure and routine, particularly significant following major life transitions. Its loss correlates with increased rumination and social withdrawal. Low independent risk; Moderate in combination with TRM signals.
VLAP does not evaluate signals in isolation. The following rules define when single Moderate or Low signals elevate to High-risk classifications requiring immediate human review.
Current validation status of each signal category and the associated NIMH SBIR Phase I research agenda. VLAP distinguishes between signals that are clinically annotated and signals that have completed powered statistical validation — we do not present the former as the latter.
VLAP's accuracy claims are grounded in active IRB-approved clinical research with a university research partner — not internal testing, not synthetic benchmarks, and not general NLP performance metrics that don't account for cultural signal specificity.
VLAP's ability to detect high-distress signals when they are present in member language — measured against clinician-adjudicated ground truth in the IRB study cohort. Sensitivity is optimized conservatively for the SHA and HOP categories, and for signal combinations that trigger High-risk escalation: we accept more false positives to minimize missed crisis signals.
Percentage of the VLAP training corpus that would be processed as unknown tokens by a standard BERT model without the extended vocabulary — representing the portion of culturally specific language that standard models are structurally unable to read.
VLAP signal accuracy is being validated through an IRB-approved clinical study with a university research partner, using production deployment data from live Vasl cohorts. The study compares VLAP signal output against clinician-adjudicated gold-standard assessments of the same member language. Results will be published in peer-reviewed literature upon study completion.
False positive and false negative rates are disaggregated across Black, Latino, LGBTQ+, first-generation, and limited English proficiency subgroups in both the training validation and the IRB study. Parity thresholds are enforced operationally — a model version that meets aggregate accuracy targets but fails subgroup parity is not deployed. Bias monitoring is an ongoing production gate, not a one-time evaluation.
Sensitivity measures how consistently VLAP detects signals when they are present — not the rate at which all surfaced signals are clinically significant in a given instance. A high-sensitivity threshold means more signals are surfaced, which is appropriate for a clinical support tool. The clinical significance of any specific signal is always determined through human clinical review, not by the model.
The active IRB study is currently in the data collection and preliminary analysis phase. Results will be published in a peer-reviewed journal upon completion. The study protocol and preliminary design documentation are available to institutional evaluators under NDA. Contact clinical@vaslhealth.com to request access.
Vasl Health's clinical validation approach is overseen by its Senior Medical Advisor, Panagis Galiatsatos, MD, MHS — Assistant Professor of Medicine at Johns Hopkins University School of Medicine. Dr. Galiatsatos provides clinical oversight on VLAP's signal detection methodology, accuracy validation approach, and non-diagnostic output framing.
VLAP is a clinical decision-support tool, not a member-facing AI. It operates entirely behind the clinical layer — invisible to the people whose language it processes. Every signal it surfaces is directed to a licensed clinician or certified coach, reviewed by a human, and responded to through human clinical judgment. The platform is built so that automated action in response to a clinical signal is architecturally impossible.
VLAP processes only language shared through Vasl's care channels — daily check-ins and coach messaging threads. Peer group posts, external social media, school email, and any other channel are not processed.
VLAP processes the language against the 47-signal taxonomy. In-memory only — verbatim text is not retained after processing. Output: a dimensional signal profile.
Coaches see a simplified surface of VLAP output in the AI Client Insights panel: plain-language pattern alerts and mood trajectory summaries for their active members. No dimensional codes, no clinical jargon.
When a member is connected to a licensed clinician, the pre-session view includes the full VLAP dimensional signal profile — dimensional codes, pattern descriptions, cultural interpretation notes, and coaching context. Accessible only to licensed clinicians.
Signals or signal combinations that meet High-risk escalation criteria are surfaced immediately to Vasl's licensed clinical supervisor team. A licensed clinician reviews the signal and determines the appropriate response. No automated action. Human judgment initiates every response.
VLAP processes the most sensitive category of user data — mental health language from youth in underserved communities. The security architecture was designed specifically for HIPAA-regulated, school-based, and community health deployment contexts. Every constraint below is architectural, not configurable.
VLAP processes member language in-memory. Verbatim input text is not stored after signal profile generation. The retained output is a dimensional signal profile — not a transcript, not a quote. This is a HIPAA technical safeguard, not a configurable setting.
HIPAA Security Rule technical safeguards implemented across all platform components — encryption in transit and at rest, access controls, audit logging, and automatic logoff. Business Associate Agreement required for all organizational deployments. Annual third-party security audit.
Annual SOC 2 Type II audit covering security, availability, and confidentiality trust service criteria. Full audit report available to institutional evaluators under NDA. Audit conducted by an independent third-party auditor.
Individual VLAP signal context is accessible only to the assigned coach (AI Client Insights summary) and the assigned licensed clinician (full dimensional profile). School staff, org administrators, and Vasl team members outside clinical supervisory functions have zero access to individual signal data — architecturally enforced.
For school district deployments, Vasl operates as a direct service provider to students. Student health data generated in Vasl is classified as health information under HIPAA — not as an education record under FERPA — and is structurally inaccessible to school administrators under any circumstances.
Population-level aggregate signal trends surfaced to org administrators use minimum cohort size enforcement to prevent de-identification by inference. Individual member contributions to aggregate data are never discernible. This constraint applies to all org-level reporting, without exception.
VLAP's clinical credibility is grounded in active institutional partnerships — not aspirational affiliations or advisory relationships that don't involve actual work. The partnerships listed below involve ongoing operational collaboration, active research, or formal clinical oversight.
An IRB-approved clinical study is in progress with an academic research partner validating VLAP's signal detection accuracy against clinician-adjudicated ground truth assessments. The study uses production deployment data from live Vasl cohorts. Results will be published in a peer-reviewed journal upon completion. The study represents the first formal independent validation of VLAP's culturally specific signal detection capabilities.
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. His advisory role involves active participation in clinical review, not nominal affiliation.
Vasl Health provides full technical documentation to qualified institutional evaluators — health systems, research institutions, school district technology teams, and health plan medical directors. All documentation is available under NDA. Contact clinical@vaslhealth.com or use the form below to initiate an evaluation request.