Deep dive into the MedIngest clinical data ingestion architecture.
MedIngest operates as a layout-agnostic, event-driven data ingestion pipeline. It converts unstructured patient records (email attachments, TIFF faxes, scanned paper files, PNG requisitions) into standardized, HL7 FHIR-compliant databases.
Preserves columns, tables, and handwritten annotations as coordinate tokens before mapping text character arrays.
Classifies medical nouns (medications, strengths, lab findings, patient demographics) and resolves clinical abbreviations.
Translates parsed names into standard clinical codes (LOINC codes for labs, RxNorm codes for drugs, and SNOMED FINDINGS).
Packs parameters into compliant US Core FHIR JSON datasets (MedicationRequest, Observation, Patient profiles).
Every document transaction returns an audit trail detailing processing durations and pipeline success statuses at each node:
# Pipeline processing transaction telemetry
{
"transactionId": "tx_abc123xyz",
"documentType": "referral_letter",
"stages": [
{ "stage": "capture", "durationMs": 140, "status": "success" },
{ "stage": "ocr_layout", "durationMs": 420, "status": "success" },
{ "stage": "clinical_ner", "durationMs": 510, "status": "success" },
{ "stage": "terminology_map", "durationMs": 120, "status": "success" },
{ "stage": "fhir_compilation", "durationMs": 80, "status": "success" }
],
"totalDurationMs": 1270
}