RESEARCH

Research you can trust.
Standards you can verify.

Every Vantyx agent, whether for mentoring or live support, operates within a framework of published, peer-reviewed research. We don't invent best practices. We implement the ones that exist.

7 research partnersยท
50+ documentsยท
Evidence-based since the 1990sยท
2 product verticals

How we source our research

๐Ÿ“„

Primary sources

We read the original papers, not summaries. MENTOR.org EEP 5th edition, SAMHSA TIP-35, EEF toolkit entries. Direct. Unfiltered.

โš™๏ธ

Ingestion pipeline

Documents are chunked, embedded via OpenAI, and stored in Pinecone. Each agent response is retrieved from this corpus, not hallucinated.

๐Ÿ”„

Continuous review

Research evolves. Our knowledge base is versioned. When MENTOR.org updates the EEP, we update the embeddings.

Mentoring

Mentoring research corpus

MENTOR.org

National Mentoring Partnership
Key Findings
  • โ†’Elements of Effective Practice (EEP) 5th Edition, the gold standard framework
  • โ†’SMART goals approach embedded across all 8 elements of quality
  • โ†’8 elements: recruitment, screening, training, matching, monitoring, support, closure, evaluation
How We Use This

Session structure, goal-setting prompts, quality measurement across every mentoring agent.

Visit MENTOR.org โ†’

SAMHSA

US Dept of Health & Human Services
Key Findings
  • โ†’Mental health first aid protocols for youth-facing practitioners
  • โ†’Crisis response protocols, TIP-35 series
  • โ†’Trauma-informed approaches improve outcomes across all demographics
How We Use This

Crisis safeguarding detection, wellness agent conversation prompts.

Visit SAMHSA โ†’

Youth.gov

US Federal Government
Key Findings
  • โ†’Positive Youth Development (PYD) framework, strengths-based, not deficit-based
  • โ†’Protective factors: caring relationships, high expectations, meaningful participation
  • โ†’Federal synthesis of 30+ years of youth development research
How We Use This

Youth agent knowledge base, resilience-building conversation strategies.

Visit Youth.gov โ†’

OJJDP / NMRC

US Dept of Justice, National Mentoring Resource Center
Key Findings
  • โ†’32% reduction in violent behaviour (Big Brothers Big Sisters study)
  • โ†’Mentored youth 55% more likely to attend university
  • โ†’Consistent mentoring reduces truancy and school exclusions
How We Use This

Impact reporting, outcome tracking, evidence for programme ROI.

Visit OJJDP / NMRC โ†’

EEF

Education Endowment Foundation (UK)
Key Findings
  • โ†’Academic mentoring effect size: +4 months of additional progress
  • โ†’Metacognition and self-regulation: +7 months
  • โ†’Targeted support for disadvantaged pupils shows strongest gains
How We Use This

Academic support agent type, evidence-backed questioning techniques.

Visit EEF โ†’

Anna Freud Centre

Children & Families Mental Health
Key Findings
  • โ†’75% of mental health conditions start before age 18
  • โ†’Schools-based support is not therapy, clear boundary definitions
  • โ†’Safeguarding in digital contexts requires explicit design, not afterthoughts
How We Use This

Safeguarding boundaries, wellbeing agent limitations, language guidelines.

Visit Anna Freud Centre โ†’

CEBM, UMass Boston

Center for Evidence-Based Mentoring
Key Findings
  • โ†’Structured sessions show 2ร— better outcomes than unstructured
  • โ†’Match quality is the single strongest predictor of success
  • โ†’Dosage matters: 12+ sessions show significantly better results
How We Use This

Agent session quality metrics, dosage recommendations, relationship quality measurement.

Visit CEBM, UMass Boston โ†’
Support

Support agent research

Customer support standards

Research on AI support response time expectations shows users expect sub-2-second responses. CSAT score benchmarks for AI agents average 4.2/5 when knowledge bases are properly curated. Escalation rates below 15% indicate healthy agent coverage.

Sources
  • โ†’Zendesk CX Trends 2024, response time benchmarks
  • โ†’Forrester Wave: AI-Powered Agents, quality scoring
  • โ†’Gartner: 80% of support orgs will use generative AI by 2025

Boundaries & compliance

AI support agents MUST NOT provide medical advice, legal counsel, or financial recommendations. We implement hard-coded refusal logic for these categories. Our approach is aligned with EU AI Act Article 6 high-risk system requirements.

Hard-coded refusals
  • โœ•Medical diagnoses or treatment suggestions
  • โœ•Legal advice or liability opinions
  • โœ•Financial investment recommendations
  • โœ•Personal data processing beyond session scope
โ€œWe believe the most dangerous AI agents are the ones with no guardrails. The second most dangerous are the ones with guardrails no one has read.โ€

Open documentation

Every research source is listed. Every framework is cited. No black boxes.

Versioned knowledge base

When research updates, our embeddings update. You always get the latest evidence.

Human escalation always available

AI agents are tools, not replacements. Human support is one click away, always.

See how it works in practice

From research to real sessions, mentoring and support.

See how it works โ†’

Ready to deploy your first agent?

Free to start. No credit card. Live in under 10 minutes.

Our knowledge is grounded in research from

NMRC / OJJDPMENTOR.orgSAMHSAYouth.govEEFAnna Freud CentreCEBMNMRC / OJJDPMENTOR.orgSAMHSAYouth.govEEFAnna Freud CentreCEBMNMRC / OJJDPMENTOR.orgSAMHSAYouth.govEEFAnna Freud CentreCEBM