Capital efficiency, operator DNA, paying healthcare pilot
SVTech is a vertical PaaS company with an unusual shape: a twenty-year commercial operator, self-funded at pre-seed, with a paying healthcare pilot, two shared-engine flagships running on Mono-PaaS, two newer expansion proof surfaces, a validated MAIA safety classifier, and a scoped Layer-6 fine-tune roadmap that builds a defensible clinical-analytics moat from accumulating signal.
Where we stand today
May 2026. No outside capital. Paying healthcare pilot. Two shared-engine flagships in active beta, plus BizFrameHub and Colombian Spanish Academy as public expansion proofs.
Founder equity, no outside capital. Entity: SVTech Consulting Services LLC, formed December 14, 2022 - three-plus years of operating history predating the AI pivot.
First paying Stage 1 pilot from a nurse-practitioner-led psychiatric practice - the same practice grown from zero to four practitioners on organic search before becoming the DWA pilot customer. Stage 2 licensee revenue follows once the engagement model graduates from custom-vertical work to platform-reviewed manifests.
DWA (HIPAA-aware, PHI) and GTM-OS (non-PHI) both run on the Mono-PaaS engine today - same identity, tenancy, AI orchestration, adapters, manifests. BizFrameHub is launching today as a 57-course AI Business Academy for SMBs and professionals. Colombian Spanish Academy is live as a rapid localized language-learning proof with 20 courses and a Bogotá guide module.
The wedge is not the whole education market. It is AI-coached vertical academies that can ship now.
HolonIQ projects education approaches $10T by 2030, and WEF/IMF data points to a real reskilling cycle. SVTech should not frame this as a generic TAM grab. The investor-relevant wedge is narrower and sharper: regulated, professional, and operator communities that need branded AI learning products live in days, not after a custom platform build.
That is where Mono-PaaS matters: your branded academy, live in days, with an AI coach that knows your curriculum, and a safety layer you can dial up to clinical grade.
HolonIQ forecasts the global education market will reach almost $10T by 2030, with workforce education among the fastest-growing segments.
WEF estimates 59 of every 100 workers will need training by 2030 as skill requirements shift.
WEF reports 85% of surveyed employers anticipate adopting workforce upskilling through 2030.
IMF analysis says nearly 40% of global jobs are exposed to AI-driven change, increasing the pressure to update skills.
Five defenses that compound
Each layer is independently valuable; together they're the reason a competitor can't replicate this in a quarter.
MAIA - continuously-retrained distress classifier
Sentinet/suicidality (ELECTRA-base, ~110M params, CC0 baseline) running as a CPU sidecar at 22.3 ms/sample. Validated F1 0.93 / crisis recall 1.0. Classifier stays text-free; retrain pipeline promotes a successor only when it measurably beats the baseline. Compliance and clinical trust enforced at the engine level via failSafe="closed" routing.
Evidence-based lesson library
922 lessons across 43 DWA courses + 974 localized English-Spanish lesson files across 49 GTM-OS courses, plus BizFrameHub's 57-course AI Business Academy and Colombian Spanish Academy's 20-course localized curriculum. Each DWA therapeutic lesson takes 15-20 hours to author at clinical quality. The broader portfolio proves both depth and repeatable packaging.
Adaptive learning · risk stratification · content-outcome correlation
Each requires longitudinal data that only emerges from real users over time. The AI layer page has the full roadmap - ClinicalBERT, RoBERTa, BKT, IRT, XGBoost, pgvector.
Practice licensing - aligned incentives
Licensed partners do the marketing to their own patient base; revenue share aligns incentives. Custom vertical → licensed deployment → (eventually) Studio self-serve. See the engagement model.
Publishable clinical outcomes
Longitudinal data produces publishable research directly. Academic credibility compounds into practice trust; practice trust compounds into distribution.
Architecture discipline as a capital-efficiency signal
One of the fastest ways to read a technical team is what they've chosen not to build. Every item on the SVTech kill list - no microservices, no Kubernetes, no custom AI gateway, no in-house vector DB, no custom auth, no multi-region by default - reduces surface area, operational cost, and hiring burden.
Every non-obvious decision is documented with context, alternatives, and consequences. docs/adrs/ is a read-the-repo-to-understand-the-company artifact.
No package lands in the shared platform without ≥2 vertical consumers or a third committed use case. Prevents premature abstractions; keeps core surface area honest.
v1 enforces tenant scoping in application code with audited query paths. The tenantLeakHarness CI gate lands alongside the v2 RLS rollout, turning tenant isolation failures into build breaks rather than incidents.
Six idempotent provision-01..06 scripts replay the full infrastructure onto an empty Dokploy. Sees-the-reality reproducibility.
Modular monolith. Root-server deployed. Redundant and predictable.
SoloFrame is being built as a modular monolith that runs on a single node today and scales to a multi-node deployment under Dokploy + Traefik + Docker Swarm - without a code fork or a rewrite. Stateless app replicas, shared state in Postgres / Redis / object storage, multi-replica-safe background jobs. Multi-node brings redundancy as well as scale: a failed node doesn't take the platform down, and rolling deploys happen without maintenance windows. Premium isolation is a topology choice, not a branch.
Deployment is to VPS and root-server infrastructure, not Vercel-style managed-PaaS hosts. That matters for AI-heavy workloads: per-invocation pricing converts engagement into variable cost. Root-server capacity is predictable; scale and redundancy come from additive nodes, not a vendor's usage meter. Platform architecture ->
Priced against what it actually unlocks
We're avoiding the "big raise" posture. Capital maps 1:1 to specific GPU-bound ML workloads, named clinical partnerships, and the next tranche of evidence.
ClinicalBERT fine-tune · RoBERTa sentiment · BKT/IRT · XGBoost risk scoring · pgvector embedding · retraining buffer. Each a scoped, independently deployable system with a clear metric.
IRB-compliant head-to-head studies with named practices and research partners. Outcome: publishable distress-detection performance + content-outcome correlation data.
Scaling the evidence-based lesson library across additional clinical tracks, with clinical-reviewer honoraria and authoring infrastructure.
Moving from founder-led licensing to a small, targeted BD effort against psychiatric networks and behavioral-health benefits providers. Operator DNA means we know what we're not hiring yet.