For organizations
For health systems & research IT

Workbench.

From certified logic to data — in your environment.

Inspector defines and certifies a scenario. The work doesn’t end there — someone still has to find that data in your warehouse and write the queries. Today that’s a data analyst spending two to three days per request, and the knowledge evaporates when the project ends. Workbench is where that stops.

01 — Two tools, one spine
Open · MIT

Inspector

Author, validate, visualize, and certify the clinical logic. Produces an audit-ready bundle. Never touches patient data. Anyone can run it.

For organizations

Workbench

Takes a certified scenario and runs it against your data— mapping each concept onto your warehouse schema, generating SQL, and remembering every mapping for the next project.

02 — The business case

Stop paying the mapping tax twice

Today a data request means an analyst hunts through the warehouse, writes the SQL by hand, and documents it — two to three days, every time. When the project ends, that knowledge walks out the door. Workbench turns that one-off labor into an asset that compounds.

Days → a session

A first cohort is scoped in a working session, not a multi-day analyst ticket.

Review, don't rewrite

High-confidence mappings auto-accept; analysts spend time only where the match is genuinely ambiguous.

It compounds

Every mapping is saved as institutional memory — the second study starts from what the first one proved.

03 — What each one does

Side by side

Capability
Inspector
open · MIT
Workbench
commercial
Author scenarios — Builder, AI, or YAML
OMOP vocabulary search (SNOMED · RxNorm · LOINC)
Validate against the PSDL spec
Decision-graph (DAG) visualization
Certified bundle + IRB Word export
Connect to your Epic EDW / data warehouse
Auto-map concepts onto your schema
Confidence-tiered analyst verification
SQL generation — Epic · OMOP · PCORnet · i2b2
Reusable institutional mapping memory
License
MIT · free
Commercial
Best for
Researchers · regulators
Health systems · research IT
04 — How Workbench works

Concept → column, with the human in the loop

For every signal in a scenario, Workbench proposes where that data lives in your warehouse — checking institutional memory first, then a synonym dictionary, then medical embeddings (FAISS) — and routes each suggestion by confidence so analysts only review what needs review.

≥ 97%Auto-accept

High-confidence matches are accepted automatically. The analyst can review, but need not act.

75 – 96%Confirm

A single suggestion is shown for a one-click confirm or reject.

60 – 74%Choose

Two or three candidate columns are offered; the analyst picks one or searches manually.

< 60%Manual map

No confident match — the analyst browses the schema with sample values and selects the column.

Every mapping is saved. The second project in your institution is faster than the first — and the tenth is nearly automatic.

05 — Where it runs

Built to clear your security team’s first question

Workbench connects to your data — so where it runs, and what leaves your network, is the question that matters. The architecture answers it up front.

In your environment

Self-hosted with Docker or Kubernetes, inside your own network. You hold the database, the keys, and the certificates. Nothing phones home.

Patient data stays put

Workbench produces dataset specs and SQL; the queries run against your warehouse, inside your perimeter. The reusable memory stores schema mappings — table and column names — not patient records.

Auditable, not a black box

Every mapping and approval is logged — who, what, when, why — and the clinical logic is the open PSDL standard: inspectable, portable, and yours to keep.

06 — What a pilot looks like

One cohort, end to end

A pilot is a working engagement, not a sales call — measured in weeks, not quarters.

01

Scope

Pick one real cohort. We map its concepts against your schema, together.

02

Verify

You review only the mappings the system is unsure about. We generate the SQL.

03

Run

You execute it in your environment and keep the dataset spec — and the mappings — for the next study.

What we need from you: schema metadata — table and column names — and an analyst partner for a few sessions. No patient data, ever — we never need access to your records. What you keep: a working dataset spec, reusable mappings, and no lock-in.

Bring PSDL to your institution

Let’s scope a pilot.

We’ll walk you through Workbench on a real scenario, look at how it maps onto your data sources, and scope a first cohort with your team. No commitment — a working conversation.