Turn unstructured medical text into structured data— on your own hardware.

ehidome reads doctor’s notes, discharge summaries and lab reports and returns diseases, drugs, dosage, frequency and PII as structured fields. It can strip patient-identifying information at the same time — and it runs locally, so medical text never has to leave your infrastructure.

Unstructured input

“Person name: John Kamau was diagnosed with Disease: Type 2 diabetes and prescribed Drug: metformin Dosage: 500mg Frequency: twice daily.”

Structured output

{
  diseases:   ["Type 2 diabetes"],
  drugs:      ["metformin"],
  dosage:     ["500mg"],
  frequency:  ["twice daily"],
  pii:        ["John Kamau"]
}

PIIDiseaseDrugDosageFrequency

Illustrative output. Extraction and de-identification run entirely on-device — no medical text is sent to a cloud service.

Use cases

De-identify medical records
Strip the 18 HIPAA Safe Harbor identifiers from clinical text, with pseudonymization, consistent date shifting, synthetic replacements and leakage auditing built in.
```
Patient:   [NAME]
Phone:     [PHONE]
MRN:       [MEDICAL_RECORD_NUMBER]
Diagnosis: hypertension
```
Extract clinical entities (NER)
Medical named-entity recognition for diseases, medications and other biomedical concepts, using purpose-built models you select by name.
```
analyze_text(
  "Patient started on imatinib for
   chronic myeloid leukemia.")

DRUG      imatinib
DISEASE   chronic myeloid leukemia
```

Structure doctor's notes

Turn assessment-and-plan free text into records: conditions with status, medications with dose and frequency — ready for your database.

{
  "conditions": [
    { "name": "hypertension",
      "status": "active" }
  ],
  "medications": [
    { "name": "amlodipine",
      "dose": "10mg",
      "frequency": "daily" }
  ]
}

Convert to FHIR
Map extracted entities onto FHIR R4 resources — Condition, MedicationStatement, Observation, Bundle — with US Core validation, SMART-on-FHIR and Bulk FHIR tooling.
```
Doctor's note
  → ehidome
  → Disease / Medication / Lab
  → FHIR resources
  → Hospital EHR
```
Assist medical coding
Once concepts are extracted, map them to the terminologies your workflows depend on: ICD-10, SNOMED CT, LOINC, RxNorm, UMLS and CMS-HCC.
```
"Type 2 diabetes mellitus"

ICD-10 candidate:  E11.9
SNOMED concept:    44054006
```
Run private, on-device AI
Ship the same pipeline everywhere: Python (CPU / CUDA), Swift via Apple MLX (ehidomeKit), Kotlin on Android with ONNX Runtime, JS with WebGPU, or REST on a server.
```
Scan discharge note
  → OCR
  → ehidomeKit
  → PII removed locally
  → Entities extracted locally
```

Turn unstructured medical text into structured data— on your own hardware.

How ehidome transforms text

De-identify medical records

Extract clinical entities (NER)

Structure doctor's notes

Convert to FHIR

Assist medical coding

Run private, on-device AI