Map, validate and monitor clinical data, automatically.
Retire fragile R-scripts and Java desktop tools. Map local terminology with AI, monitor your OMOP and custom-CDM pipelines continuously, and run every computation inside your own VPC — patient rows never leave.
# secure agent started inside local VPC
✓ control-plane endpoint connected
$ omopstack init-omop-scan --source prod_clinical
# compiling 3,450 OHDSI v6.0 conformance rules…
plausibility 99.8% ok
conformance 100.0% standard
$ omopstack ai-map --col local_drug_name --std RxNorm
"HCT 25mg / Lisinopril 20mg Tab" → RxNorm 316866
✓ auto-assigned · confidence 0.98
2,451 variables mapped — no raw rows moved
Compatible with global clinical architectures
Clinical data teams are fighting their own tools.
Mapping, quality and ETL still run on single-user desktop apps and hand-run R scripts — slow, siloed, and stale the moment a vocabulary updates. Here's the day-to-day, side by side.
Hours to rebuild the vocabulary index
Usagi rebuilds a local Lucene index from Athena files — on every machine.
Single-user desktop, files emailed around
No collaboration, no versioning, no audit trail.
Quality checks run by hand, then go stale
DQD & Achilles in R → static HTML, out of date within days.
Your laptop, or a DIY Docker stack
Ops burden on your team; PHI moving where it shouldn't.
Hosted, pre-indexed vocabulary
Map in minutes with AI semantic search — nothing to build locally.
Multi-user, versioned, governed
Git-style review & approval with a full audit trail.
Continuous quality, trended & alerting
3,500+ checks always-on, with Slack / PagerDuty alerts.
Managed, inside your VPC
Zero PHI egress. SOC 2 · HIPAA · BAA.
Terminology mapping is where OMOP projects stall. OMOPStack removes the bottleneck.
Three modules, one observability plane.
Replace a sprawl of legacy desktop tools and R-scripts with a single secure, collaborative platform built for biostatistics and clinical-AI pipelines.
Usagi Cloud Mapping
Move past single-user desktop mapping. Build clinical dictionary maps collaboratively and let embedding models parse local shortcodes to SNOMED CT, RxNorm and LOINC.
- Collaborative multi-user validation
- Semantic pre-mapping & re-rank
- Git-style change approval
DQD Live Observability
Automate database quality without hand-running R. Continuously execute 3,500+ OHDSI-compliant data-health checks inside your own database engine, with history and alerting.
- Schema-drift auto-detection
- Slack, PagerDuty & Teams alerts
- Historic quality trend tracking
Multi-Model Synthesis
Transform, audit and clean data flowing from vendors' OMOP schemas into your proprietary custom models — without schema drift or lost clinical context.
- Dual-ended transform checking
- Custom CDM mapping interfaces
- Native transformation logging
From raw source to monitored OMOP — one flow.
A single VPC-resident agent profiles, maps, validates and monitors. No desktop apps, no data leaving your boundary.
Connect
Point the VPC agent at your source database — outbound-only, no ports opened.
Profile
Continuous scan of schema, distributions, keys and drift — re-runs on every refresh.
Map with AI
Semantic automapping to OMOP standard concepts, scored — reviewed, never fabricated.
Validate
3,500+ OHDSI quality checks across conformance, completeness and plausibility.
Monitor
Always-on observability: trended over time, with alerts when thresholds break.
Every capability, in depth.
Browse the full feature set by area.
Dual-channel retrieval
Lexical (trigram) + semantic (BioLORD) candidate generation over the full vocabulary.
Constrained LLM re-rank
Ranks only real concepts — structurally cannot fabricate a concept_id.
Calibrated confidence
Per-domain auto-accept thresholds; everything else routed to review.
Review queue
Top-k candidates with evidence and rationale — approve, flag or reassign.
Git-style governance
Versioned change sets, approval workflow and a full audit trail.
Active learning
Corrections fine-tune the model — your maps get better the more you use it.
3,500+ OHDSI checks
The full Kahn framework: conformance, completeness and plausibility.
Continuous & scheduled
Runs on every data refresh — not by hand, not once a quarter.
Trending & drift
Time-series history so you catch regressions the moment they appear.
Alerting
Slack, PagerDuty and Teams notifications when thresholds break.
Drilldown
Per-table, per-field and per-concept violation detail.
OHDSI export
Standards-compliant JSON results for your own records.
Source profiling
Types, distributions, candidate / surrogate keys and dedup statistics.
Visual + AI mapping
Source→target arrows, AI-suggested and collaboratively reviewed.
Runnable ETL
Emits executable SQL / dbt — design and execution in one place.
OMOP 5.3 / 5.4 / 6.0
Targets the current CDM versions out of the box.
Custom CDM
Map to your own proprietary model — not just OMOP.
Vendor synthesis
Transform external OMOP feeds into your custom schema, cleanly.
Runs in your VPC
BYOC or on-prem; the computation engine never leaves your boundary.
Zero PHI egress
Only aggregate statistics ever reach the control plane.
SOC 2 · HIPAA · BAA
Audit-ready, with ready-to-sign business associate agreements.
SSO & RBAC
SAML / OpenID, role-based access and true multi-tenancy.
Outbound-only
No inbound ports; no database exposed to the internet.
Air-gap capable
Deploys into fully isolated, disconnected environments.
One platform for the stack you're stitching together.
OMOPStack consolidates the OHDSI mapping, quality and ETL tools — most of them desktop, single-user, and run by hand — into one collaborative platform that runs continuously inside your VPC. It integrates with the analytics tools your team already trusts.
Usagi Cloud Mapping
Replaces
Java desktop · single-user · string-matching · hours to build a local index
→ AI semantic automapping with confidence scores and human-in-the-loop review — collaborative, governed, and grounded in the standard vocabulary (no fabricated concept_ids).
DQD Live Observability
Replaces
Manual R runs · static HTML · point-in-time · no trending or alerting
→ The 3,500+ OHDSI checks run continuously, trended over time with alerting and multi-user triage — the Kahn framework, hosted and always-on.
Multi-Model Synthesis
Replaces
Desktop arrow-drawing · emits docs, not runnable ETL · OMOP-rigid
→ AI-assisted source→target mapping that profiles continuously and emits executable ETL — for OMOP and your own custom CDM.
We connect to the analytics and infrastructure your team already trusts. No rip-and-replace of validated study methodology.
Full coverage — every tool, mapped
| Tool | Category | OMOPStack |
|---|---|---|
| Usagi | Terminology | Replaces |
| IMO · Symedical · Apelon | Terminology (commercial) | Competes |
| Hecate · Lettuce | Semantic vocab search | Competes |
| DataQualityDashboard | Data quality | Replaces |
| Achilles · ARES | Characterization | Replaces |
| Rabbit-in-a-Hat · WhiteRabbit | Structural ETL | Replaces |
| Perseus | Web ETL | Competes |
| ATLAS · WebAPI · HADES | Analytics & evidence | Integrates |
| Strategus · Broadsea | Orchestration & infra | Integrates |
| Athena · OMOP vocabulary | Vocabulary source | Built on |
Replaces — we supersede it · Competes — modern tool in the same space · Integrates / Built on — we connect to it or anchor to it, never reinventing validated methodology or fabricating concept_ids.
Try OMOPStack in real time.
Interact with live simulations of our two core products.
Pick a messy clinical entry or write your own. The AI parses the text and matches it to the vocabulary using context.
No active term mapped
Select or enter a phrase on the left to run the engine.
Deciphered input
"HCTZ 25mg + Lisinopril 20mg Tab"
98% MatchMatches in standard vocabulary
Mock metrics from an OMOP database of 10,000,000 patient records. Filter by OHDSI category.
| OHDSI rule | Table | Violation | Status |
|---|---|---|---|
| IsPlausibleBeforeDeath | DRUG_EXPOSURE | 0.00% · 0 / 10M | PASS |
| CheckFK_Person_Id | CONDITION_OCC | 4.12% · 412k / 10M | WARN |
| CheckNotNull_Gender | PERSON | 0.01% · 1k / 10M | PASS |
| IsPlausibleAgeUnder120 | PERSON | 0.00% · 4 / 10M | PASS |
| InvalidConceptIdsExist | MEASUREMENT | 11.89% · 1.19M / 10M | FAIL |
Patient data never leaves your VPC perimeter.
A zero-trust BYOC / on-premise framework: the control plane and analytics UI are hosted by us, while the computation engine runs entirely inside your private infrastructure.
Zero ingestion of PHI
The agent pushes only aggregate counts (e.g. 5 misses / 10M rows). Patient-level data stays put.
No inbound firewall openings
The local daemon makes outbound connections only — no exposed database ports.
Audit-ready compliance
Passes HIPAA, SOC 2 Type II and GDPR audits. Ready-to-sign BAAs.
OMOPStack SaaS
Control plane
- · Multi-user web portal
- · Analytics aggregates
- · Alert orchestrator
Client VPC subnet
Data plane
- · OMOPStack computation engine
- · Internal clinical databases
- · Raw rows remain here
What is manual monitoring costing you?
Estimate your platform subscription and the engineering value you recapture.
Listed on AWS & Google Cloud Marketplaces — apply committed cloud budgets to bypass procurement.
Recommended deployment
AI-assisted terminology mapping for up to 8 tunnels.
Questions about clinical observability.
Usagi and DQD are excellent community tools but are desktop-based (large Java downloads, huge index files) and run manually via R scripts. OMOPStack turns those algorithms into a modern multi-user web app: collaborative terminology mapping with cloud models, and continuous pipeline-quality monitoring.
Yes — that's the core architecture. The agent is a lightweight docker container inside your private subnets. It talks only to your database, computes locally, and reports only metadata and health percentages. No patient-level records or PHI ever leave your VPC.
We provide full OHDSI OMOP CDM support out of the box, but the core is schema-agnostic. Load custom database maps via simple JSON/YAML to use the AI mapping interface and validation rules across any legacy healthcare model.
Rather than string-matching, it uses clinical semantic embedding models. Given a local description, it evaluates synonyms, abbreviations and target-vocabulary hierarchies to compute similarity scores against SNOMED, LOINC and RxNorm — with constrained re-ranking so no concept_id is fabricated.
Retire the desktop OHDSI stack.
Map terminology with AI, monitor data quality continuously, and deploy in your VPC within days — patient rows never leave your boundary.