scene 1 · 0:00-0:08 · problem + product
Nebius 2026 AI Discovery Award · Digital Health

Digital Wellness Academy

A HIPAA-aware clinical mental-health education platform, in active beta deployment - curriculum researched against current clinical literature using deep research and clinically validated by board-certified psychiatric nurse practitioners at Real Psychiatric Services, protected by the MAIA distress-classification safety architecture.

Behavioral-health licensees - psychiatric practices, multi-clinic franchises, behavioral-health benefits providers, university counseling centers - operate DWA as their credentialed digital-education arm.

SVTech Consulting Services LLC · A 90-second technical walkthrough · Every claim verifiable in the source

Product  digitalwellness.academy Entity  svtechconsultingservices.com Sister vertical  solofounder.academy
scene 2 · 0:08-0:50 · live workflow
Screen before escalate. One alert per episode. Production-grade.

MAIA classifies. The coach screens. The clinician responds.

Every AI ingress is gated by MAIA - sentinet/suicidality, ELECTRA-base, C-SSRS clinical training. On the first euphemism we do not fire an alert. The coach enters screening mode and asks one open-ended calibration question. On heavier disclosure the alert fires - one row, one episode, even if MAIA fires three times across screening turns. The provider sees a pre-computed payload: anonymized alias, transcript with MAIA classifications inline, urgency tier, lesson suggestions frozen at fire-time, PMHNP-curated response templates.

Patient · /coach digitalwellness.academy/coach
I'm tired
Wellness Coach · MAIA: none (0.12)
That's understandable. Tiredness can come from a lot of different places. Want to tell me a bit about what's going on - is this physical fatigue, or more of a heavy feeling that won't lift?
I don't understand this lesson
Wellness Coach · MAIA: none (0.08)
That's okay - some of these concepts take a couple of passes to land. Which lesson are you in?
Cited: Depression: From Understanding to Action / Lesson 3 - Getting Moving Again (Behavioral Activation Basics) · evidence: STRONG
I don't know if it's worth it anymore
Wellness Coach · screening mode MAIA: mild (0.71)
When you say "worth it" - can you tell me more about what you mean? I want to make sure I hear you right. There's no wrong answer.
I mean my life
Wellness Coach · crisis surface MAIA: crisis (0.91)
I hear you - and I want you to know your feelings are real and you don't have to carry this alone. Please call or text 988 right now to reach the Suicide & Crisis Lifeline. They're trained for exactly this moment, available 24/7. I've also flagged this to your care team so a clinician can reach you.
Mike Sullivan, Provider CLINICIAN
I just got the alert and I'm here. Please call or text 988 right now if you haven't yet - they're trained for exactly this moment. To reach me directly, call our office at 555-555-1234 and give code AB34X7 (active for 7 days). The "Getting Moving Again" lesson in your Depression: From Understanding to Action course has a behavioral activation grounding technique some patients find helpful between calls - lesson 3, evidence grade STRONG, authored by Dr. [PMHNP]. I'm here. - Mike
$ psql -c "SELECT * FROM ai_classification_event ORDER BY created_at DESC LIMIT 1"
 
event_id: 3a8c7e1f-... · context: coaching
classifier: distress · model_id: sentinet/suicidality · model_version: v1.0
level: crisis · confidence: 0.913 · processing_ms: 22
screening_mode: true · alert_fired: true · episode_id: 8c2f1a-...
Provider · /provider/alerts digitalwellness.academy/provider/alerts
Critical
Patient: "Riverbend" "I mean my life" · MAIA crisis (0.91) · screening: 1 turn
just now
Elevated
Patient: "Larkspur" 2 screening turns, ambiguous · MAIA mild→mild
2h ago
Monitoring
Patient: "Cedar" MAIA mild once · de-escalated in screening
yesterday
Resolved
Patient: "Marigold" Sent appointment code · acknowledged 12h later
3d ago
Alert detail · Patient "Riverbend" Critical

Triggering excerpt (last 4 turns + MAIA)

Patient none 0.12
I'm tired
Patient none 0.08
I don't understand this lesson
Patient mild 0.71
I don't know if it's worth it anymore
Patient crisis 0.91
I mean my life

Pre-computed lesson suggestions (frozen at fire-time)

Depression: From Understanding to Action · Lesson 3 - Getting Moving Again
55 min · Behavioral Activation (Lewinsohn) · evidence STRONG · fit 0.91
"Behavioral activation candidate; turns reference low motivation + curriculum confusion."
Depression: From Understanding to Action · Lesson 5 - Challenging Negative Thoughts
55 min · CBT cognitive restructuring · evidence STRONG · fit 0.79
"Cognitive distortion pattern in 'is it worth it' framing."

PMHNP-curated response templates · Critical tier

Office urgent-call code with 988 reinforcement
"I just got the alert and I'm here. Please call 988 right now... To reach me directly, call {{office_phone}} and give code {{appointment_code}}..."
PMHNP-curated · ships with platform · evidence-flagged
988 first-response (no office bridge)
"Please call or text 988 immediately. They're trained for exactly this moment..."
PMHNP-curated · ships with platform
Compassionate stabilization with course reference
"Your feelings are real. While we connect, here's a grounding technique from..."
PMHNP-curated · ships with platform

Architecture: screening state machine (deriveState + decide as pure functions) in lib/safety/episode.ts · alert dedup keyed on (patient, session) in lib/safety/session-store.ts · lesson suggestions from retrieveLessonsForCoach in lib/ai/rag.ts · 7 PMHNP-curated response templates seeded at boot · production-grade data model and state machine; alert detail composer is demo-grade UI on production endpoints; patient-side care_message rendering ships in open beta. Full architecture in docs/positioning/screening-and-alert-architecture.md.

scene 3 · 0:50-1:05 · why Nebius matters
The promotion gate, empirically

Our first SVTech-built distress fine-tune was rejected by the promotion gate. Sentinet remains production.

A weaker gate would have promoted v1 on its in-distribution F1 alone and shipped a worse-than-baseline experience. Our gate caught it. Nebius compute is what feeds the gate the production-class candidates that earn it - LLaMA-3-8B coach LLM, multilingual MAIA, per-tenant fine-tunes.

v1: Rejected

SVTech distress v1

DistilBERT-base, 67M params. Trained on 5,000 stratified thePixel42 samples in 4.7 minutes on Apple Silicon M2 Pro MPS, 2 epochs.
In-distribution F1: 0.929 · Adversarial recall: 0.647 (vs sentinet 0.765) · Trap_FP: 2 (vs sentinet 0)
Sentinet: Production

sentinet/suicidality

ELECTRA-base, ~110M params. C-SSRS clinical training. CC0. Three independent validation surfaces, all reproducible from services/maia/validation/.
Synthetic crisis_recall: 1.000 · Adversarial crisis_recall: 0.682 [95% CI 0.59-0.76] · External corpus F1: 0.870

The promotion gate runs on Apple Silicon today. Nebius compute scales the gate's candidates - the canonical Layer-6 in-house coach LLM, multilingual MAIA retraining cadence, per-tenant fine-tune capacity.

scene 4 · 1:05-1:20 · measured outcome
Six-layer cost architecture · Reproducible from vendor pricing

$0.594 → $0.0955 today. $0.0438 with Layer 6.

Per 20-turn coaching session. Same output, two architectures. Reproducible from unit-economics-model.csv using vendor-published pricing - re-runnable in any audit. The DWA wellness-session counterpart runs at 4.4x today; the Layer-6 ambitious projection assumes the in-house coach LLM Nebius unlocks absorbs 60% of heavy reasoning.

Naive (frontier-only)
$0.594
baseline
Sonnet on every turn. The pre-architecture cost.
SoloFrame today
$0.0955
6.2× measured
Layers 1-5 in production: tier routing, multi-provider via OpenRouter, prompt + context caching, RAG (pgvector + RRF for GTM; JSONB + pure-JS cosine for DWA), CPU sidecar classifiers (MAIA).
With Layer 6 (Nebius unlocks)
$0.0438
13.6× projected
In-house coach LLM (LLaMA-3-8B class) absorbs 60% of heavy reasoning. The canonical Layer-6 artifact and the largest single cost-architecture lever the platform has.
scene 5 · 1:20-1:30 · the ask
Self-funded · Bootstrapped · Eligible

The repo is the receipt.

SVTech Consulting Services LLC requests Nebius compute credits to deliver Layer 6 in production, scale MAIA retraining cadence, and unlock multilingual coverage for the documented English-only limitation. DWA in active beta at digitalwellness.academy. Open beta May 2026 acquires the marquee at-scale Stage-1 beta partner. Worldwide-when-localized: NHS, EU AI Act, APAC equivalents are manifest variants, not codebase forks.

What Nebius credits unlock

  • Item #5 Option A · course-recommender fine-tune. Smallest training run; validates the data-export → train → eval → promotion pipeline on a low-risk surface. 3-4 days of GPU time once preconditions land.
  • Item #5 Option B · in-house coach LLM (LLaMA-3-8B class). The canonical Layer 6 artifact. Conservative 6.7× / ambitious 13.6×. Largest single cost-architecture lever.
  • MAIA retrain cadence at production scale. Multilingual MAIA retraining closes the documented English-only limitation.
  • Per-tenant fine-tune capacity. Multi-tenant Layer 6 at production volume.

Contact

Mike Sullivan, Founder
SVTech Consulting Services LLC
github.com/SoloFrameHub/Mono-PaaS
private; available on request
Supplementary materials: EXECUTIVE_SUMMARY.md nebius-application-draft.md screening-and-alert-architecture.md manifest-contrast.md unit-economics-model.csv comparable-winners-analysis.md MAIA MODEL_CARD.md FINETUNE_V1_FINDINGS.md FINETUNE_COACH_V01_FINDINGS.md ADVERSARIAL_200_FINDINGS.md what-this-does-not-do.md compliance-readiness.md