The platform operates as a continuous empirical research cycle. It helps psychologists transform their observations into structured data, combining classical methods of evidence-based psychology with the analytical capabilities of AI to design modern care standards.
Powered by the Prevention Intelligence Engine (PIE) — a constraint system that keeps generative AI within scientifically validated psychological boundaries, guaranteeing its safety and strict adherence to professional protocols.
Prevention-AI is designed on a Copilot principle. Artificial intelligence never makes diagnoses, never makes decisions for the specialist, and does not replace a human. It serves as an expert assistant, relying on strict methodological frameworks embedded in the database: WHO guidelines, Functional Behavioral Assessment (FBA) protocols, and Cognitive Behavioral Therapy (CBT) approaches.
The system does not allow "free fantasizing" by algorithms. Any specialist request is processed through a strict state matrix, guaranteeing the validity of recommendations.
Determines the exact operational phase of the intervention cycle:
X1_Problem — Acute tension identificationX2_Diag — Non-clinical preventive screening / structured checkupX3_Goal — Setting limited de-escalation milestonesX4_Action — Deployment of micro-action protocolsX5_Eval — Automated reflection and outcome trackingEstablishes the safety perimeter and determines routing thresholds:
Y1_Normal — Standard age-related crisisY2_Risk — Accumulation of subclinical risk factorsY3_Problem — Manifested behavioral or communicative crisisY4_Crisis_Clinical — Hard boundary for warm routing to a specialistCategorizes the physiological or ecological origin of the stress vector:
M1_Biology — Age-related and neurodevelopmental shiftsM2_Psychophysiology — Somatic markers and stress responsesM3_Cognition — Internal behavioral cycles, beliefs, and scriptsM4_Social — Microsystem dynamics (family and peers)M5_Environment — Macrospaces (institutional and digital contexts)Dynamic context filtering based on the certified user interface: Psychologist (methodological journals), Teacher (classroom dynamics), or Administrator (macro reports).
Determines the target sociological system size for the data cycle: Individual → Family → Group → Community → Society.
To ensure full semantic continuity between Tier 1 consumer apps and Tier 2 specialist spaces, each session generates an isolated, PII-free cryptographic passport. This passport tracks behavioral change dynamics (movement from baseline anomaly to target de-escalation) and evaluates context continuity in longitudinal logs, without revealing personal stories.
Instead of treating nodes as simple links, the system maps connections as a semantic galaxy of 438 structural nodes and over 1,800 directed edges. This topological structure serves as a catalog of safe pathways, guiding the specialist along developmental, clinical, and preventive tracks.
We utilize a fully transparent, highly deterministic prompt hierarchy. It establishes non-clinical boundaries, isolates sensitive diagnostics, and safely guides the LLM along validated protocols. This entire structure is open for collaborative governance and ongoing scientific review.
Generates all main types of psychologist documents from raw data, structures according to methodology, provides recommendations, and formats into a print-ready document.
Compiles Functional Behavioral Assessment (FBA) and psychological profiles, mapping student behavior to A-B-C triggers.
Analyzes roles and calculates hidden social and destructive risks.
Automatically transforms unstructured voice dictation into strict Data, Assessment, and Plan (DAP) fields for professional journals.
Safely handles acute intake scenarios (e.g., adoption panic) and maps conversational taxonomy to provider psychological profiles.
Post-processing of test results to generate a single empathetic "deep dive" question and synthesize PII-stripped evidence summaries.
Manages relationship support logic (family or couple). Ensures absolute neutrality, privacy protection, and deflects security threats away from the AI.
Translates party positions through the lens of Nonviolent Communication (NVC), helping the parent see the teen's hidden needs, and the teen understand the adult's anxiety without disclosing private chat details.
Decomposes dense incoming messages with multiple complaints into 2–4 independent themes. Identifies priority sprint focus and formulates one micro-action step for the next 24 hours.
Tracks therapeutic contact depth and switches response logic (Discover → Summarize → Mobilize), separating facts from parental interpretations and offering ready-made letter/request templates.
Absolute unshakable baseline: prohibits clinical diagnoses, moralizing, and mandates routing to emergency services (e.g., 911) when life-threatening data is inputted.
In addition to the macro-research loop, Prevention-AI already utilizes a closed operational cycle in every paid session. Raw chat is never saved as evidence; the factory outputs structured keys, protocol IDs, sprint metadata, and user ratings that can be aggregated to determine protocol efficacy.
Each user utterance is bound to taxonomy coordinates: problem code, category code, severity (Y-axis), modality cues (M-axis), and prevention vector. Unknown themes are flagged for review without saving message text.
The engine selects bounded micro-interventions from the protocol catalog for the current stage (X1–X5). Guidelines are injected into the AI runtime; applied protocols are logged at every assistant step.
Dialogue is organized as short case sprints: tension identification → focal problem exploration → escalation containment → evaluation → next step decision. Each sprint contains a unique ID, focus node, and session stage, so the same taxonomy language covers all utterances and sessions.
When a sprint closes, the user evaluates the cycle: helped / partially / didn't help / got worse, plus optional scientific scales (clarity, relief, confidence in next step, alliance, micro-action status). Outcomes are linked to applied protocols and focal taxonomy, measuring what worked without retaining narratives.
From Telemetry to Evidence: Instead of collecting raw personal chats, the platform captures only anonymized event signals (such as classified problem categories, protocol steps, and user-rated sprint outcomes). These aggregated data points generate real-time statistical maps, showing which prevention protocol works best for a specific problem. This approach automatically builds a comprehensive, high-fidelity statistical pool for public health decision-makers — something traditional paper reporting systems could never achieve.
The platform moves beyond classical static questionnaires. By analyzing natural conversation, we create a continuous background diagnostic loop that evaluates risk factors without disrupting the therapeutic flow.
Immediate dialogue initiation via dynamic tagging, replacing static forms with real-time mapping of structural themes (e.g., REL_FAM, DEV_EMO).
Decomposition of validated psychometric scales (depression, anxiety, PTSD) into atomic elements, enriched with IRT (Item Response Theory) weights, LLM system prompts, and contextual guardrails.
A background AI module that parses free text and automatically maps client narratives against diagnostic scales without explicit questions, governed by strict context filtering.
Computerized Adaptive Testing where the LLM organically weaves the next most statistically valuable question from the test bank into the natural flow of conversation.
An interactive dashboard for specialists displaying screening probabilities (but not clinical diagnoses) and anonymized conversational evidence, plus a test builder for researchers.
Collection of anonymized statistics to compare AI shadow scoring against traditional forms, calibrating IRT weights for continuous accuracy improvement.
Local client-side logging (IndexedDB) and edge PII stripping are design choices ensuring individual interactions can become structured research signals rather than just passive charts, feeding into the evidence cycle described above.
The target architecture pushes behavioral typologies and risk vector matrices into analytics warehouses (e.g., BigQuery) so institutions can track regional stress patterns at an aggregate level. This governance layer is not yet operating at a national scale; institutional pilots aim to validate the pipeline end-to-end in localized contexts.
We leverage high-performance vector databases (Azure AI Search / Vertex AI Vector Search) to perform deterministic matching between natural language input and our fixed prevention ontologies, forcing the model to stay on script.
Local PII (names, specific locations) never reaches the LLM cloud API. The edge application strips identifiers before generating embeddings, guaranteeing HIPAA and GDPR compliance for any subsequent macro-analysis.
"This is how empirical science, cutting-edge preventive practice, and regional governance will finally operate in a single, real-time data feedback loop. By turning everyday field activity into a structured, verifiable source of evidence, we aren't just applying existing protocols — we are establishing the future standards of AI-assisted psychological care."
