Healthcare data interoperability

Transform HL7, FHIR, and X12 with AI, without running PHI through one.

Describe the mapping and bring your spec. Data Shepherd writes a tested, reviewable transformation script during the build phase. After that, every run executes the sealed script in an isolated sandbox. No patient data is sent to an AI model at run-time.

Free plan available. No card required.

No PHI to a model at run-time · build from a spec alone if you cannot share data
oru_sample.deid.hl7 → results.csv
Input · HL7 v2 (synthetic, de-identified)
MSH|^~\&|LAB|HOSP|EHR|CLINIC|202606051200||ORU^R01|0001|P|2.5
PID|1||MRN-000123^^^HOSP||DOE^JANE||19800101|F
OBX|1|NM|GLU^Glucose||96|mg/dL|70-99|N
▾ AI-written transformation, reviewed and versioned
Output · analytics-ready
patient_idtestvalueflag
MRN-000123Glucose96 mg/dLNormal
MRN-000124Glucose112 mg/dLHigh
Runs with no AI. PHI stays out of the model.

The problem

Your team is already pasting PHI into ChatGPT.

When a mapping is due and the integration queue is backed up, someone drops a patient file into a chatbot to reshape it quickly. It works, and it is a HIPAA incident waiting to happen. Your team needs the speed without the exposure.

40%

of workplace AI interactions involve sensitive data, often through unmanaged personal accounts.

Source: Cyberhaven

How it works

From a messy interface file to a tested pipeline, in minutes.

No integration engineer, no six-week mapping project.

1

Describe it and bring your spec

Plain English plus your implementation guide, companion guide, or data dictionary. A de-identified sample sharpens the result. Cannot share data at all? Build from the spec alone.

2

AI writes transparent scripts

You get reviewable, versioned Python, never a black box. Inspect every line before it touches a record.

3

Run it anywhere, with no AI

Trigger by API, schedule, or click, or pull from SFTP, S3, or Azure Blob. The sealed script does the work; no PHI goes to a model.

The AI builds it. A sealed script runs it.

Most AI data tools push records through a language model on every run. We keep the model in the build phase only, so production never sends patient data to a model.

Build (one time): AI writes the scripts, from your spec, or a de-identified sample you control
Every run after: sealed sandbox, no network, no AI, no PHI to any model

Full data-flow table, sub-processor list, and retention schedule in our Trust Center →

Speaks your formatsHL7 v2FHIR R4X12 837 · 835270 · 271834 · 820EDIFACTCSV · Excel · fixed-width

Built for healthcare workflows

The transformations your team does by hand today.

837 → CSV

Claims normalization

Flatten and validate 837P and 837I claims into a clean schema for your clearinghouse or analytics.

835 → AR

Remittance posting

Turn 835 ERA files into postable rows for your AR system, with adjustment codes mapped.

270/271

Eligibility flattening

Convert eligibility responses into a row-per-member table your ops team can actually read.

HL7 → FHIR

ADT to FHIR R4

Map HL7 v2 ADT messages to FHIR resources, or whatever shape your EHR or downstream system expects.

ORU → table

Lab results to analytics

Parse ORU^R01 results into an analytics-ready table with units and reference flags.

834 recon

Enrollment reconciliation

Diff 834 enrollment files against your roster to surface adds, terms, and changes.

Security and compliance

What your security team needs, by default.

No PHI to AI at run-time

Guaranteed by architecture

HIPAA-ready

BAA program · in progress

SOC 2 Type II

in progress

Full audit trail

Every run and change logged

Stop choosing between fast with AI and HIPAA-safe.

Map a sample HL7 or X12 message in minutes, and hand your security team a data-flow table instead of a risk-acceptance form.