Nebius 2026 AI Discovery Award - Digital Health semifinalist

Nebius judges: start with Digital Wellness Academy.

SVTech Consulting Services LLC is the corporate applicant and the SoloFrame / Mono-PaaS home. Digital Wellness Academy is the healthcare product entered for the award.

If you arrived from the SVTech domain, this page is the handoff. The proof packet lives on DWA because that is where the MAIA safety architecture, clinical content system, data flywheel, and Nebius compute ask are documented.

Start with the video

Watch the DWA clinical workflow first.

Provider notes enter the portal, the system runs a privacy scan, recommends courses, sends the student a learning-path link, and later supports a student-owned progress share link.

Onboarding
Dashboard
Lessons
Which site is which

SVTech is the company. DWA is the award entry.

The application points at SVTech because the LLC, platform architecture, and Mono-PaaS thesis live here. The evidence the committee needs first is the DWA technical packet.

Corporate applicant

SVTech Consulting Services LLC

The legal entity, founder story, and platform company behind SoloFrame / Mono-PaaS.

Digital health entry

Digital Wellness Academy

The clinical education product with MAIA safety, provider workflows, 922 lessons, 21 assessments, and the data flywheel.

Compute ask

Nebius unlock

Production-class in-house coach LLM, faster MAIA retraining, multilingual coverage, and per-tenant fine-tunes.

DWA evidence addendum

The next proof layer is clinical workflow intelligence.

This section summarizes the RAG, clinical-notes, and patient-compliance-report architecture: the same curriculum, RAG, MAIA, provider portal, and audit patterns extend into higher-value clinical workflows without turning the LLM into a clinician.

RAG intelligence layer

The curriculum becomes runtime knowledge.

DWA already grounds coaching and provider-facing search in its own evidence-linked content corpus. This section specifies the next layer: hybrid retrieval, reranking, and per-surface retrieval rules across coach, onboarding, provider prep, and clinical workflow surfaces.

Provider workflow

Clinical notes point to learning paths.

The proposed notes-to-learning-path workflow keeps the clinician in control: de-identify the note, extract clinical themes, map those themes to DWA lessons through RAG, then present a draft path for provider review and sharing.

Engagement evidence

Between-session work becomes observable.

The patient compliance report spec turns completed lessons, assessments, thought records, tracking logs, and coaching interactions into a patient-controlled report scoped to a date range instead of a vague engagement claim.

Why this improves the Nebius application

Technical sophistication

The answer moves from "AI coach plus distress classifier" to a multi-surface intelligence system: curriculum-grounded RAG, MAIA gating, provider search, notes-to-learning-path drafts, and engagement reporting.

Real-world impact

The loop becomes visible: a clinician can assign evidence-linked between-session work, and the patient can bring back a dated record of lessons, assessments, thought records, logs, and coaching engagement.

Nebius infrastructure need

The compute story becomes sustained production workload, not a one-time training ask: MAIA retraining, coach-model work, continuous embedding, reranking, PHI-safe redaction/generation chains, and batch report generation.

Conservative judge shorthand

DWA closes the loop between clinician guidance and patient practice: provider-led notes-to-learning-path drafts, patient-controlled engagement reports, and curriculum-grounded RAG running inside a PHI-safe multi-model architecture. Because compliance, safety gates, and RAG are baked into the engine, new AI capabilities can be composed from existing building blocks in hours instead of quarters.

Boundary for judges

This addendum does not claim autonomous treatment planning or diagnosis. The architecture keeps content knowledge in RAG so the curriculum can change without retraining, reserves fine-tuning for MAIA and future coach-model work, and leaves clinical decisions with the provider.

Global reach boundary

Platform globalization is proven. Clinical globalization is planned and gated.

DWA is English-first today because clinical localization is full-market healthcare work. It requires local clinician review, jurisdiction-specific crisis resources, consent language, and jurisdiction-specific compliance posture.

The platform underneath has already proven its first GTM-OS localizations through the LatAm (Spanish) and Brazil (Portuguese, just shipped) market surfaces: same core engine, localized buyer personas, market-specific workflows, regional messaging patterns, local integration choices, and no platform fork. That is a repeatable GTM-OS localization pattern, not the only geography. DWA follows the same architecture globally, with a higher clinical validation bar in each market.

Platform proof today
DWA globalization path
First GTM-OS localizations: LatAm (ES) + Brazil (PT)
Global DWA curriculum localization
Same Mono-PaaS engine
Market-specific clinician review
Localized workflows, personas, and messaging
Jurisdiction-specific crisis resources
No forked core
MAIA validation across target languages
Commercial vertical already localized
Compliance manifests per jurisdiction
What this does not claim

SVTech does not claim that localized DWA markets are clinically ready today, that MAIA is validated across target languages, that jurisdiction-specific crisis workflows are implemented, or that jurisdiction-specific clinical compliance has shipped. The honest line is: localization without forking is proven at the platform layer; global clinical localization launches only after local review, crisis routing, consent, compliance, and model-validation gates clear.

Why this is not LLM-wrapper slop

Three things separate a platform from a Claude-shell-with-prompts.

Where AI is bounded out. How research enters the build. What the gate looks like that says no. Each item below maps to a published framework, not a vibe.

01 · BoundedOWASP LLM06 · NIST 600-1 G-9

AI bounded OUT of the crisis path.

MAIA classifies signals. A deterministic state machine with 7 PMHNP-curated templates generates every crisis response. No LLM ever writes the safety reply. Excessive-agency risk is closed at the architecture layer, not patched at the prompt layer.

02 · GatedNIST AI RMF MS 2.1

Promotion gate has already rejected models.

A model can clear in-distribution accuracy and still fail real-world spot-check. That has happened here. Receipts live in the repo. Strong benchmark numbers do not ship to users until the human gate signs off.

03 · CitedISO 42001 A.5 · A.7

Research tiers gate every clinical claim.

Four-tier evidence hierarchy. Tier-1 (peer-reviewed) and Tier-2 (clinical guideline) are required to ground any shipped clinical or behavioral claim. Vendor blogs sit at Tier-4 and cannot back a production assertion. Research briefs expire on a clock.

04 · DefendedOWASP LLM01 · LLM08 · LLM10

Three-layer prompt-injection defense, CI-enforced.

Input sanitization, delimiter hardening, output validation. Tenant-leak harness in CI. PHI-aware logging redacts when compliance.phi: true. Per-conversation cost tracking with automatic model swap at $0.15.

Frameworks aligned (not certified · alignment + readiness)
NIST AI RMF 1.0 + 600-1OWASP LLM Top 10 2025OWASP Agentic Top 10 2026ISO/IEC 42001:2023 (AIMS)NIST SSDF v1.2MITRE ATLASCSA AISMM (target L3)G7 SBOM-for-AISLSA L2 buildsEU AI Act readiness

Quarterly governance review against the CSA AI Security Maturity Model. SBOM + signed releases on every production artifact. AI model inventory + provider allowlist version-controlled in repo.

Committee proof signals

The technical packet answers the skeptical questions first.

The strongest evidence is not a pitch claim. It is shipped architecture, measured cost control, deterministic safety gates, and a product surface running on the same platform thesis described here.

Solo-founder throughput

1 founder. 922 lessons. 0 new hires.

DWA shows agentic AI used as an operating system for production output, not as a demo layer. The architecture scales the output. The org chart does not have to.

Infrastructure moat

Compliance is a flag, not a fork.

DWA and GTM-OS run on the same SoloFrame / Mono-PaaS engine with different manifests: PHI-aware MAIA routing for DWA, noop classifier for GTM-OS.

Unit economics

$0.594 to $0.0955 per 20-turn session.

The six-layer cost architecture produces a 6.2x reduction today. Nebius compute unlocks Layer 6: the in-house coach LLM and per-tenant fine-tunes.

Safety discipline

42 state-machine tests. Zero LLM calls in CI.

The MAIA promotion gate is deterministic enough to reject a model that misses the bar, with classifier-before-LLM routing and fail-closed behavior.

Recommended review path

For Nebius evaluation, start with the DWA proof map. Use SVTech for corporate context, SoloFrame / Mono-PaaS architecture, and the portfolio-wide platform thesis.

Go to the DWA Nebius proof map ->

New on the DWA packet: a six-criteria scorecard and the global-impact case.

New on Digital Wellness Academy · June 1, 2026
Platform velocity proof

Neuroplasticity School (Mind & Performance) ships on schedule

A new evidence-graded school added to DWA on the existing engine - manifest-driven, no routing or engine changes. 17 courses · ~225 lessons across five pillars, each lesson carrying an explicit evidence grade (Strong / Moderate / Emerging) and embedded into the same RAG corpus MAIA uses for "why does this work?" coaching. It is the Mono-PaaS thesis in one shipment: a new school drops in as content, not a fork.

Brain Science FoundationsMindset & Cognitive RewiringMindfulness & AttentionBody–Brain PerformanceIdentity & Purpose
Read the whitepaper →DWA proof packet →
New · Fourth production vertical shipped

60-Day Founder - a new vertical on the same engine.

The engine that runs MAIA-served clinical wellness also runs commercial GTM, SMB AI education, and now an AI startup academy - one inference and orchestration stack, four manifest-declared verticals, compliance as a flag not a fork. 60-Day Founder shipped as content on the same infrastructure the Nebius proof packet describes.

257 lessons · 32 courses · 6 tracksAI build coach8 artifact workshopsclassifier: noop · non-PHI
See the 60-Day Founder vertical →