Skip to content

Ask AI

The Ask AI section explains how to configure and use the AI assistant in CareFlow Kids.

Unlike the AI clinical support page inside the Sick Visit documentation (which explains how AI is used during a consultation), this section explains how the AI system itself is configured and controlled by the physician.

AI integration in CareFlow Kids follows three main principles:

  • Physician‑controlled configuration
  • Structured clinical input (JSON)
  • Clear separation between local guideline rules and external AI analysis

AI is optional and can be disabled simply by leaving the configuration empty.

AI architecture overview

The AI system operates using the following pipeline:

flowchart LR
    A[Structured visit data] --> B[Episode JSON snapshot]
    B --> C[Physician prompt template]
    C --> D[Online AI provider]
    D --> E[Structured AI answer]

The physician controls both:

  • which AI provider is used
  • how the AI is instructed to answer (prompt rules)

AI providers

CareFlow Kids supports external AI providers. However this feature is still experimental at the moment and has been mainly tested with OpenAi's ChatGPT, which means connection with other providers such as Google's Gemini and Anthropic's Claude may not work as expected.

Typical configuration includes:

  • API endpoint (for example https://api.openai.com/v1)
  • Model name (for example gpt‑5‑mini)
  • Provider selection (OpenAI or compatible provider)
  • API key

These settings are configured in the Physician profile → AI assistant section.

The API key is stored locally and used only when the physician explicitly requests AI assistance.

No AI request is sent automatically.

Even when AI is configured, the system only sends clinical data to the AI provider when the physician explicitly clicks Ask AI. If the feature is not used, no data is transmitted.

Prompt configuration

CareFlow Kids allows the physician to define custom prompt templates.

Prompts control:

  • how the AI interprets clinical data
  • what sources it should rely on
  • the format of the response

The physician defines how the AI assistant behaves using a configurable prompt template available in the Physician profile section of the application.

Typical prompt instructions include:

  • using trusted pediatric medical sources
  • structuring answers in a predictable format
  • highlighting evidence quality
  • generating coding suggestions when appropriate

Prompts can also instruct the AI to generate useful clinical outputs such as:

  • structured visit summaries
  • parent information handouts
  • medical certificates
  • patient instructions or care guidance

Because the AI receives structured clinical information, these outputs can be generated in a consistent and clinically relevant format.

Because prompts are physician‑editable, the AI assistant can be adapted to different practice styles.

JSON preview

CareFlow Kids generates a structured JSON representation of the episode before sending an AI request. JSON (JavaScript Object Notation) is a standardized text format used to organize structured data.

The JSON preview feature allows the physician to inspect the exact information that will be transmitted to the AI provider.

This improves transparency and makes debugging AI behavior easier.

The JSON typically contains:

  • patient age and sex
  • symptoms and complaints
  • abnormal vital signs
  • examination findings
  • problem listing
  • relevant contextual information
  • relevant patient history (for example perinatal history or past medical history when available)

Direct identifiers such as name, alias, or MRN are never included.

Advanced usage

Advanced users may adapt the AI configuration to suit specific workflows.

Examples include:

  • refining prompts to emphasize certain medical sources
  • adjusting response structure
  • customizing coding instructions
  • experimenting with different AI models

Because the AI assistant always receives structured clinical data, prompt changes can significantly influence the quality and usefulness of the responses.

Clinical responsibility

AI responses are advisory only.

They may help with:

  • summarizing information
  • exploring differential diagnoses
  • suggesting investigations

However, clinical decisions remain the responsibility of the physician.

The AI assistant is designed to support — not replace — clinical judgement.