Care Entry Workflow

A caregiver finishes a patient visit and dictates a short voice note describing what they did during the visit. Your nursing-home information system turns it into a structured care entry with identified interventions, administered medications and consumables, ready for billing and patient records.

The workflow at a glance

┌──────────────┐    ┌────────────┐    ┌──────────┐
│ Caregiver    │    │ Your CIS   │    │ xander   │
└──────┬───────┘    └─────┬──────┘    └─────┬────┘
       │ voice note       │                 │
       │─────────────────>│                 │
       │                  │ 1. transcribe   │
       │                  │────────────────>│
       │                  │ 2. care report  │
       │                  │    + classify   │
       │                  │────────────────>│
       │                  │ 3. SELECT FROM  │
       │                  │    catalog WHERE│
       │                  │    cat IN (...) │
       │                  │ (your DB)       │
       │                  │ 4a. interventions
       │                  │ 4b. medications │
       │                  │ 4c. consumables │
       │                  │ (parallel)      │
       │                  │────────────────>│
       │                  │   3 results     │
       │                  │<────────────────│
       │                  │ persist care    │
       │                  │ entry           │

Step 1: Transcribe the voice note

Same call as Field Service step 1. Use POST /prompt/audioTranscription for short notes (under ~20 minutes). Set the language explicitly for better domain-specific recognition (medical vocabulary).

curl -X POST "$XANDER_BASE_URL/prompt/audioTranscription" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "AudioInput": {
      "AudioFile": {
        "Base64AudioFile": "<base64-encoded audio bytes>",
        "FileName": "visit-2026-04-11-0930.m4a",
        "AudioType": "audio/mp4"
      }
    },
    "Options": {
      "PromptIdentification": {
        "ProjectId": "your-project-id",
        "Promptname": "care-entry/audio_transcription"
      },
      "SessionMetadata": {
        "UseCaseName": "care-entry-workflow",
        "UseCaseSessionId": "<visit id from your CIS>"
      }
    },
    "TranscriptionOptions": {
      "AudioLanguage":  { "LanguageCodeIso639": "de" },
      "TargetLanguage": { "LanguageCodeIso639": "de" }
    }
  }'

Tip: domain dictionary

For domains with specialised vocabulary (drug names, diagnostic terms), pass a QueryParameter named dictionary with a list of terms the model should be aware of. This dramatically improves the transcription accuracy of proper names.

"QueryParameter": [
  { "Name": "dictionary", "Value": "Furosemid, Pantoprazol, Metformin, Ramipril" }
]

Step 2: Generate a care report and identify coarse categories

Use POST /prompt/informationextraction with a free-form prompt to generate a Markdown care report and a list of category hints (e.g. "vital-signs check", "medication administration", "wound care"). You'll use these to filter your catalogues in step 3.

curl -X POST "$XANDER_BASE_URL/prompt/informationextraction" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "UserQuery": {
      "UserQueryText": "<the transcript from step 1>"
    },
    "QueryParameter": [],
    "Options": {
      "PromptIdentification": {
        "ProjectId": "your-project-id",
        "Promptname": "care-entry/care_report_summary"
      },
      "SessionMetadata": {
        "UseCaseName": "care-entry-workflow",
        "UseCaseSessionId": "<same visit id>"
      }
    }
  }'

Step 3: Reduce your catalogues (in your CIS)

Filter your interventions, medications and materials catalogues by the patient's care plan, the categories from step 2, and any business rules you have. Aim for a few dozen candidates per category.

SELECT articleNumber, name, unit
FROM intervention_catalog
WHERE category IN (...) AND wardCode = $patient.ward;

SELECT sysId, name, dose, unit
FROM medication_catalog
WHERE id IN (SELECT medicationId FROM patient_prescriptions WHERE patientId = $patient.id);

SELECT articleNumber, name
FROM consumable_catalog
WHERE category IN (...);

Step 4: Identify interventions, medications and consumables (in parallel)

Three parallel POST /prompt/informationextraction-structured calls. Each gets the transcript plus the relevant reduced catalogue and a JSON schema describing the response shape.

Step 4a: Identify interventions

curl -X POST "$XANDER_BASE_URL/prompt/informationextraction-structured" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "UserQuery": { "UserQueryText": "" },
    "QueryParameter": [
      { "Name": "Leistungskatalog",   "Value": "<reduced interventions catalogue>" },
      { "Name": "Leistungsbeschrieb", "Value": "<the transcript from step 1>" }
    ],
    "Options": {
      "PromptIdentification": {
        "ProjectId": "your-project-id",
        "Promptname": "care-entry/identify_interventions"
      },
      "SessionMetadata": {
        "UseCaseName": "care-entry-workflow",
        "UseCaseSessionId": "<same visit id>"
      }
    },
    "OutputStructure": {
      "JsonSchemaDefinition": "{\"type\":\"object\",\"properties\":{\"interventions\":{\"type\":\"array\",\"items\":{\"$ref\":\"#/$defs/Intervention\"}}},\"required\":[\"interventions\"],\"additionalProperties\":false,\"$defs\":{\"Intervention\":{\"type\":\"object\",\"properties\":{\"articleNumber\":{\"type\":\"string\"},\"name\":{\"type\":\"string\"},\"reason\":{\"type\":\"string\"},\"confidence\":{\"type\":\"number\",\"description\":\"1=low, 5=medium, 10=high\"},\"quantity\":{\"type\":\"number\"}},\"required\":[\"articleNumber\",\"name\",\"reason\",\"confidence\",\"quantity\"],\"additionalProperties\":false}}}"
    }
  }'

Step 4b: Identify medications

Same shape, with the medication catalogue. Note: medication identification is the most safety-critical of the three. Use confidence-score filtering aggressively in your UI and never auto-bill medications below a high confidence threshold without human review.

"QueryParameter": [
  { "Name": "Leistungsbeschrieb",   "Value": "<the transcript from step 1>" },
  { "Name": "Medikamentenkatalog",  "Value": "<reduced medications catalogue>" }
]

Step 4c: Identify consumables / materials

"QueryParameter": [
  { "Name": "Leistungsbeschrieb", "Value": "<the transcript from step 1>" },
  { "Name": "Materialkatalog",    "Value": "<reduced consumables catalogue>" }
]

Gotchas and tips

• Patient PII in correlation IDs. Use opaque IDs (your CIS visit ID), not patient names or social insurance numbers, in UseCaseSessionId. The correlation ID lands in our internal logs and we don't need PII there.
• Pre-filter medications by patient prescription list. The single biggest accuracy gain in this workflow is constraining the medication catalogue to what the patient is actually prescribed. A patient on 8 medications produces much better results than the entire ward's pharmacopoeia.
• Confidence thresholds for billing. A confidence score of 9-10 is typically safe to auto-bill; 6-8 should go to a human reviewer; below 6 is a strong signal that the intervention or medication was either not actually performed or that your catalogue is missing the right entry.
• Prompt provisioning. The Promptname values shown (care-entry/identify_interventions etc.) are placeholders. Your xander contact will agree the actual template names with you when they provision the prompts for your tenant.

Next

For document-based workflows (delivery notes, prescriptions, lab reports, scanned forms) instead of voice, see PDF Document Extraction. For long handover or shift-change recordings, see Meeting Helper and the async transcription pattern.