Products & Labs · Clinical platform
Verum
Clinical inference platform · payer + provider
The verdict is deterministic. The AI is everywhere else. A clinical-inference platform where verdicts are computed by code, and LLMs handle extraction, narrative and second-opinion texture.


Verum · v.1.x · liveOverview / Clinical platform

The thesis
If extraction is correct, the verdict is correct by construction.
Verum makes a deliberate split between load-bearing and decorative AI. Extraction is genuinely a language task and stays LLM-bound. The verdict is a code path: rule-tree evaluation in pure TypeScript — same answer across every model, every provider, every run.
Two products run on that one engine. Verum · Decide handles payer-side prior authorization; Verum · Insight handles provider-side multimodal clinical decision support. The BRIDGE connects them at the data layer, so a prior-auth reviewer can pull a vision read from Insight into a case mid-review.
- Deterministic gate
- Verum Vision · 4B clinical model
- Cross-payer fan-out
- Append-only audit
- FHIR R4 export
- SOC 2 Type II
For: Payer utilization-management teams · provider clinical-decision groups
Discuss this productCapabilities
Two production surfaces. One engine.
Two production-grade clinical-decision surfaces on one deterministic-gate engine: Decide for payer prior auth, Insight for provider multimodal decision support.
Prior authorization, reviewer-native
Four procedure families encoded across Medicare, Commercial and Medicaid, with 2024 and 2026 policy versions replayable. Policy is encoded as data, not prompt.
Cross-payer fan-out
The same clinical extraction runs three lanes in parallel, each against its own policy ontology. Every cited rule traces back to verbatim policy text.
Multimodal decision support
A six-stage pipeline — extract, annotate, image, fuse, gate, synthesize — across chest X-ray, diabetic retinopathy, dermatology and prostate pathology flagships.
Vision in the critical path, never the verdict
A proprietary 4-billion-parameter clinical vision model contributes structured facts. It never fabricates findings on the wrong modality — it returns confidence 0 instead.
Append-only ledger, sealed records
Every action since a case opened is on the ledger. Sealed records are immutable, and every sealed record exports as a PDF decision packet and a FHIR R4 bundle.
Resolve, override, counterfactual
Reviewers resolve missing facts, override with an advisor verdict, or probe what would flip a decision — each action captured next to the verdict it changed.
Model-portable by construction
Five providers, forty-eight models, one switch to swap live. Customer keys are threaded per request and never persisted server-side.
How it works
From clinical note to auditable verdict.
The agent pipeline behind every decision — probabilistic where it should be, deterministic where it must be.
- 01
Clinical note
Free-text, structured or voice input arrives with the case.
- 02
Extracted facts
LLM extraction emits typed, source-quoted fact spans against a per-flagship schema.
- 03
Deterministic gate
Rules × facts → verdict, in pure code. No temperature, no sampling — same facts, same answer.
- 04
Decision + audit
APPROVE · PEND · DENY, with the cited rule, counterfactual and a sealed audit record.
- 2
- Production products on one engine
- 4B
- Parameter proprietary clinical vision model
- 5
- LLM providers · 48 models
- 22
- Built-in walkthrough tours
Compliance & audit posture
- SOC 2 Type II
- FHIR R4 export
- Append-only audit ledger
- Deployable in customer-tenant VPC
Discover more
More from the Insignia portfolio.
Bring Verum into your operation