# CureSelect AI — Real-Time Voice Intake & EHR Frontend Integration Guide

**Version**: `2.0.0 (Production Release)`  
**Target Audience**: Frontend Engineers (React, Next.js, Vue, React Native, Flutter, Vanilla TypeScript)  
**Backend Protocol**: Full-Duplex WebSockets (`ws://`) & HTTP/1.1 REST  
**Authentication Header**: `X-API-Key: skyheal_secure_api_key_2026` or `Authorization: Bearer <TOKEN>`  
**Default Dashboard**: `http://localhost:8000/` or `http://localhost:8000/dashboard`  

---

## 1. Architectural Overview

The **CureSelect AI Voice Engine** is a high-speed, full-duplex conversational health intake system. It processes real-time microphone audio or text turns, classifies speaker perspectives, detects emergency distress, checks RxNorm drug safety, redacts HIPAA/ABDM PII, and generates real-time HL7 FHIR R4 Bundles and Doctor SOAP Notes.

```mermaid
sequenceDiagram
    participant Client as Frontend (React / Web / Mobile)
    participant WS as WebSocket Manager (ws/auto)
    participant Engine as AI Orchestrator & Multi-Models
    participant EHR as EHR Export (FHIR / SOAP Note)

    Client->>WS: Connect (ws://localhost:8000/v1/voice-onboarding/ws/auto)
    WS-->>Client: {"type": "connected", "session_id": "VOICE-AUTO-8F12A9C4"}
    WS-->>Client: {"type": "ai_response", "reply": "Hello Patient, welcome to CureSelect..."}
    
    Client->>WS: Stream Binary Mic Bytes (MediaRecorder WebM/PCM)
    WS-->>Client: {"type": "backchannel", "reply": "Mm-hmm..."} (< 15ms)
    WS-->>Client: {"type": "speculative_audio_chunk", "ttfa_ms": 180} (< 200ms)
    WS-->>Client: {"type": "transcript", "speaker_type": "caregiver", "speaker_label": "Caregiver (Mother)"}
    WS-->>Client: {"type": "ai_response", "reply": "I'm sorry to hear about your mother's headache...", "audio_base64": "..."}

    Client->>EHR: GET /v1/voice-onboarding/soap-note/{session_id}?format=html
    EHR-->>Client: Render Printable Doctor SOAP Note Report
```

---

## 2. WebSocket Connection Lifecycle (`ws://localhost:8000/v1/voice-onboarding/ws/auto`)

### 2.1 Connection Setup
Connect to the zero-effort auto-session WebSocket endpoint:
```typescript
const ws = new WebSocket("ws://localhost:8000/v1/voice-onboarding/ws/auto");
```

### 2.2 Inbound WebSocket Event Schema

#### 1. Connection Initialized (`connected`)
Sent immediately upon socket handshake:
```json
{
  "type": "connected",
  "session_id": "VOICE-AUTO-9DEC79EC",
  "message": "Connected to CureSelect Real-Time Voice Engine"
}
```

#### 2. Speaker Diarized Transcript (`transcript`)
Sent as soon as patient or caregiver speech is recognized:
```json
{
  "type": "transcript",
  "role": "patient",
  "speaker_type": "caregiver",
  "speaker_label": "Caregiver (Mother)",
  "text": "My mother has severe headache for 10 days and she is taking Metformin 500mg."
}
```
*Note*: `speaker_type` is either `"patient"` or `"caregiver"`.

#### 3. Low-Latency Acoustic Backchannel (`backchannel`)
Emitted in **< 15ms** during patient pauses to signal active listening:
```json
{
  "type": "backchannel",
  "reply": "Mm-hmm...",
  "audio_base64": "..."
}
```

#### 4. Speculative Audio Chunk (`speculative_audio_chunk`)
Emitted in **< 200ms** on the first 3-4 words generated by the LLM:
```json
{
  "type": "speculative_audio_chunk",
  "clause_text": "I am sorry to hear that.",
  "audio_base64": "...",
  "ttfa_ms": 180
}
```

#### 5. Complete AI Response & State Update (`ai_response`)
Contains full reply, TTS audio, next stage, and updated clinical state:
```json
{
  "type": "ai_response",
  "reply": "Any known drug allergies?",
  "audio_id": "aud_771829",
  "audio_base64": "...",
  "next_stage": "symptom_collection",
  "completed": false,
  "state_updates": {
    "symptoms": [{"name": "headache", "duration": "10 days", "severity": "severe"}],
    "medications": [{"name": "Metformin", "dosage": "500mg daily"}]
  }
}
```

#### 6. Emergency Nurse Escalation Alert (`nurse_escalation_alert`)
Emitted when Pain Score >= 8/10 or Distress Score > 0.75:
```json
{
  "type": "nurse_escalation_alert",
  "session_id": "VOICE-AUTO-9DEC79EC",
  "distress_score": 0.85,
  "pain_score": 9,
  "message": "CRITICAL: Patient reported severe pain (9/10). Nurse escalation triggered!"
}
```

#### 7. RxNorm / FDA Drug Safety Warning (`drug_safety_warning`)
Emitted when drug interactions or drug-allergy contraindications are flagged:
```json
{
  "type": "drug_safety_warning",
  "warnings": [
    {
      "severity": "CRITICAL",
      "category": "DRUG_INTERACTION",
      "description": "High risk of bleeding: Combined use of Warfarin and Aspirin."
    }
  ]
}
```

---

## 3. Outbound Client Messages

### 3.1 Streaming Audio Bytes
Stream raw WebM / PCM audio blobs from `MediaRecorder`:
```javascript
mediaRecorder.ondataavailable = (e) => {
  if (e.data.size > 0 && ws.readyState === WebSocket.OPEN) {
    ws.send(e.data); // Send raw Blob
  }
};
```

### 3.2 Text Turn Fallback
```json
{
  "type": "text_turn",
  "text": "10 days-ah severe headache iruku, Metformin 500mg sapudren."
}
```

### 3.3 Full-Duplex Barge-In Interruption
Send when the user speaks or clicks "Interrupt AI" while AI audio is playing:
```json
{
  "type": "interrupt"
}
```

---

## 4. REST API Integration Endpoints

All REST requests require the authentication header:  
`X-API-Key: skyheal_secure_api_key_2026`

### 4.1 HL7 FHIR R4 Bundle Export
- **Endpoint**: `GET /v1/voice-onboarding/fhir/{session_id}`
- **Response**: Production-grade HL7 FHIR R4 Bundle (`Patient`, `Encounter`, `Condition`, `MedicationStatement`, `AllergyIntolerance`).

```bash
curl -X GET "http://localhost:8000/v1/voice-onboarding/fhir/VOICE-AUTO-9DEC79EC" \
     -H "X-API-Key: skyheal_secure_api_key_2026"
```

### 4.2 Doctor SOAP Note & Printable PDF/HTML Report
- **Endpoint**: `GET /v1/voice-onboarding/soap-note/{session_id}?format=json` (or `format=html`)
- **Query Params**:
  - `format=json`: Returns structured Subjective, Objective, Assessment (ICD-10-CM & SNOMED-CT codes), and Plan JSON.
  - `format=html`: Renders a clean printable HTML document for 1-click doctor chart printing.

```bash
# Get JSON SOAP Note with ICD-10 & SNOMED CT
curl -X GET "http://localhost:8000/v1/voice-onboarding/soap-note/VOICE-AUTO-9DEC79EC?format=json" \
     -H "X-API-Key: skyheal_secure_api_key_2026"

# Get Printable HTML Report
curl -X GET "http://localhost:8000/v1/voice-onboarding/soap-note/VOICE-AUTO-9DEC79EC?format=html" \
     -H "X-API-Key: skyheal_secure_api_key_2026"
```

---

## 5. React / Next.js Custom Hook Example

```typescript
import { useState, useEffect, useRef } from "react";

export function useVoiceIntakeWebSocket(url: string = "ws://localhost:8000/v1/voice-onboarding/ws/auto") {
  const [sessionId, setSessionId] = useState<string>("");
  const [messages, setMessages] = useState<Array<any>>([]);
  const [clinicalState, setClinicalState] = useState<any>({});
  const [escalationAlert, setEscalationAlert] = useState<string | null>(null);
  const [drugWarnings, setDrugWarnings] = useState<Array<any>>([]);
  const wsRef = useRef<WebSocket | null>(null);

  useEffect(() => {
    const ws = new WebSocket(url);
    wsRef.current = ws;

    ws.onmessage = (event) => {
      const data = JSON.parse(event.data);
      if (data.type === "connected") {
        setSessionId(data.session_id);
      } else if (data.type === "transcript") {
        setMessages(prev => [...prev, { role: data.role, text: data.text, label: data.speaker_label }]);
      } else if (data.type === "ai_response") {
        setMessages(prev => [...prev, { role: "assistant", text: data.reply, label: "AI Assistant" }]);
        if (data.audio_base64) {
          const audio = new Audio("data:audio/wav;base64," + data.audio_base64);
          audio.play().catch(console.error);
        }
        if (data.state_updates) {
          setClinicalState(data.state_updates);
        }
      } else if (data.type === "nurse_escalation_alert") {
        setEscalationAlert(data.message);
      } else if (data.type === "drug_safety_warning") {
        setDrugWarnings(data.warnings);
      }
    };

    return () => ws.close();
  }, [url]);

  const sendTextTurn = (text: string) => {
    if (wsRef.current?.readyState === WebSocket.OPEN) {
      wsRef.current.send(JSON.stringify({ type: "text_turn", text }));
    }
  };

  const interruptAI = () => {
    if (wsRef.current?.readyState === WebSocket.OPEN) {
      wsRef.current.send(JSON.stringify({ type: "interrupt" }));
    }
  };

  return { sessionId, messages, clinicalState, escalationAlert, drugWarnings, sendTextTurn, interruptAI };
}
```