Four products. Five layers. One audit trail.

Veridata is a single substrate expressed as four products (Connect, Compass, Insights, Discovery) sitting on top of a five-layer stack. Each layer is independently auditable. Each product shares the same semantic model and the same permission model.

First design partners: leading US diagnostic labs · oncology networks · population-scale programs

Architecture

A stack you can explain to a regulator.

Read the stack as a chain of contracts: a question at Layer 05 binds down through plan, semantic model, and data plane, all the way to a trust boundary at Layer 01. Every hop is signed. Every surface traces back to the raw source.

Layer 05User surface · Connect, Compass, Insights, DiscoverySigned client sessions
Layer 04Agent & reasoning · plans, citations, attestationsTyped pipeline DSL · tool use · evidence grading
Layer 03Semantic model · clinical ontologyOMOP · FHIR R4 · HL7 v2 · DICOM SR · VCF
Layer 02Data plane · query, storage, lineageRow-level security · column encryption · materialized lineage graph
Layer 01Trust substrate · identity, audit, isolationBYOK via KMS · VPC isolation · append-only audit log · RBAC

Methodology Every signed artifact ships with three things: the query plan that produced it, the data-plane state hash at render time, and the signer's identity. Six months later, a regulator asking "how was this decided?" can re-run the same plan and verify the same hash. This is what we mean by provenance by construction.

Interaction layer

The architecture surfaces through Composer.

A single natural-language interface lets clinicians, analysts, and AI agents interact with Compass, Insights, and Discovery. Same substrate, same lineage, same trust contract — entered through the question you would have asked anyway.

01 · Gateway

Connect.

Bidirectional clinical gateway. Multi-protocol inbound across EHR, LIS, LIMS, and NGS systems; outbound delivery to EDCs, FHIR endpoints, and downstream analytics. Tenant-isolated, retry-aware, observable end-to-end — every write idempotent, every failure recoverable, every message traceable to its source.

  • 01Receive: HL7 v2.x over MLLP, FHIR R4, VCF, REST webhooks, manual upload. Canonical parsers normalize lab observations, genomic variants, and order messages into a single schema on arrival.
  • 02Harmonize: LOINC-aware lab parsing, UCUM unit normalization, terminology mapping. Field-level transforms are versioned, lookups are auditable, and every record carries provenance back to its source message.
  • 03Deliver: EDC dispatcher with adapters for Veeva Vault, Medidata Rave, Oracle Clinical One, and generic FHIR-EDC. Per-adapter retry. Per-adapter circuit breaker. Failed writes land in a DLQ and replay from the command center.
Veridata Connect — Command Center view: 7 active channels (Element AVITI VCF inbound, LabCorp HL7, Epic FHIR pull, Guardant360 outbound, FoundationOne, NeoGenomics, Veridata→Vault EDC sandbox), 2,022 messages/24h, 99.6% success, DLQ 4, summary panel showing AVITI Sequencer source mappings and Epic FHIR R4 delivery with safe-harbor de-id.
Connect · Data planeReceive · Harmonize · Deliver
Veridata Compass — Composer mid-flight: top 15 gene variant counts, classification table with AMP tier and actionability, generated SQL, and handoff buttons for trial matching and stage breakdown.
Compass · Clinical intelligencePlan · Execute · Attest
02 · Analyze

Compass.

Deterministic clinical intelligence enriched by real-time biomedical knowledge. Variant classification, therapy matching, trial screening, evidence assembly. Each step reproducible and signed, with cognitive reasoning when the evidence requires it.

Cognitive reasoning activates only on borderline cases, with deterministic fallback guaranteed.
ClassificationAMP/ASCO/CAPTier I–IV with evidence
Trial catalogLiveClinicalTrials.gov · site-local
Knowledge16+ and countingCIViC · ClinVar · OncoKB

Real-time enrichment via three-tier resolution: your database first, live API second, curated baseline always available. Every connector result tagged with provenance.

Composer is how clinicians and analysts talk to Compass. Natural language in, deterministic typed plan out. Same engine, different entry point.

03 · Answer

Insights.

Ask in plain English through Composer. Cohort discovery, revenue-cycle analytics, screening gaps, trial performance — each answer rendered from the same evidence graph that drives Compass, backed by row-level lineage.

An Insights artifact is not a screenshot. It is a snapshot of the underlying query, the data-plane state at render time, and the signer's identity. Reproducible six months later, audit-ready same afternoon.

Veridata Insights — ctDNA surveillance answer rendered from a plain-English prompt: 64,375 patients, 156,458 measurements, 4,506 with rising signal. Signal classification donut, MED status bars, and per-signal breakdown table with VAF stats. Generated SQL exposed below.
Insights · Clinical analyticsSurveillance · trends · signed
Veridata Insights — MGUS population screening rendered from a plain-English prompt. 26,132 total patients across 11 cancer types, 9,261 actionable. Bar chart by cancer type and stage; breakdown table with patient count, average age, female/male split, and actionable counts.
Insights · Population screeningPopulation gaps · cancer-type breakdown · signed
Pattern library32 patternsacross 12 clinical + operational domains
WarehousesBigQuery · SnowflakeDialect-aware SQL
ExportsPPTX · XLSX · FHIR · CSVSigned at point of export
04 · Discover

Discovery.

The drug discovery substrate for translational teams: variant effect prediction, organoid drug response, digital pathology, population-scale reference atlases. Built on the same engine as Compass and Insights, with complete provenance from sample to insight.

Veridata Discovery — Eleanor Chen digital pathology review. H&E Diagnostic Slide from TCGA-LUAD lung adenocarcinoma, whole slide image with molecular markers including EGFR L858R, MET amplification, TP53 R273H. Imaging metadata tied to molecular findings with trial matching.
Discovery · Digital pathologyWSI · TCGA-LUAD · molecular-linked
Veridata Discovery — Theo Jensen variant effect prediction. NVIDIA Evo2 7B pathogenicity scores for BRAF V600E, TP53 Y220C, and TERT promoter mutations. ESMFold structural impact predictions with 3D protein structure visualization for druggable cavity detection.
Discovery · Variant effectEvo2 · ESMFold · 3D structure
Variant effect
Frontier models.

Evo2 for DNA pathogenicity scoring. ESMFold for protein structure, pLDDT confidence, and druggable cavity detection.

Functional
Organoid to therapy.

Drug screen ingestion, IC50 curve fitting, response classification, toxicity signals, therapy prediction. Gene-drug maps across oncology, cardiac, hepatic, renal.

Reference
Research-grade atlases.

Cancer Model Passports for organoid metadata, ChEMBL for drug activity, Cellosaurus for cross-references. Harmonized into a single query surface.

Provenance
Signed results.

Every research query returns a signed, reproducible artifact. Same trust contract as Compass and Insights.

Consumption modes

Four ways to consume it.

Every way to consume Veridata runs on the same deterministic execution engine. Labs, developers, integrators, and AI agents all call the same pipeline steps, with the same lineage, from the same engine. Most platforms ship one of these. Few ship all four on one substrate.

Surface · 01
Platform UI

Browser-based console for configuring pipelines, managing connectors, running analytics, reviewing executions. Composer natural-language interface. Admin and end-user modes.

For: lab operators · clinical teams

Schedule a demo →
Surface · 02
Python SDK

pip install veridata-sdk. Author custom pipelines against the same deterministic engine. @step decorator, hash-chain provenance, compliance-grade lineage built in.

For: bioinformaticians · partner devs

Request docs access →
Surface · 03
HTTP API

Direct HTTP access to the platform: clinical pipelines, evidence packaging, FHIR push, signed report generation. Authenticated, rate-limited, with full OpenAPI spec.

For: integrators · platform engineers

Request API reference →
Surface · 04
MCP Server

Every clinical AI calls Veridata as a standard tool. Not an integration project. A protocol. Evidence from 16+ connected sources, clinical reports with full provenance.

For: Claude · GPT · Gemini · any MCP client

Request early access →
Compliance posture

Built for clinical-grade.

A snapshot of the architectural posture the platform is built on.

PrivacyHIPAAPHI · BAAs supported
Regulated records21 CFR Part 11 readyElectronic signatures · audit
SigningCryptographicHash-chain provenance
EncryptionAES-256 · BYOKCustomer KMS supported
IsolationPer-tenant VPCRow-level security
Audit logAppend-onlyEvery action, every actor
Deployment

Your cloud. Your VPC. Your control.

Veridata runs in your cloud, in your VPC, under your security policies. No data leaves your infrastructure.

Containerized
Cloud-native by design.

Fully containerized architecture that scales to zero when idle and scales to demand under load. Same platform, same APIs, same compliance guarantees regardless of where it runs.

Docker · Kubernetes · PostgreSQL · BigQuery · Snowflake

Your infrastructure
Deploy where you run.

Veridata stands up in your cloud project, your VPC, under your security policies. Cloud-agnostic by design. Initial deployments on GCP; AWS, Azure, and on-premise available on request.

GCP · AWS · Azure · On-premise

Next

See it working against your data.

See the demo → Email the team