Phase 2 — Voice AI
The core of KaiVox — an AI receptionist that answers phone calls, converses with callers using GPT-4o, books appointments, detects sentiment, and logs every call with a transcript.
What It Does
When someone calls your business phone number (Vobiz or Twilio), KaiVox answers in under 3 seconds. The AI speaks your greeting, listens to the caller, understands their intent, responds naturally, and can book appointments in real time. When the call ends, the full transcript is saved, sentiment is analysed, and automation rules can trigger.
Key Routes
| URL | Description |
|---|---|
POST /api/vobiz/{slug}/answer | Vobiz call answer — returns greeting XML |
POST /api/vobiz/{slug}/input | Vobiz speech input — processes caller speech via OpenAI |
POST /api/vobiz/{slug}/hangup | Vobiz hangup — logs call, runs sentiment, fires automations |
POST /api/twilio/{slug}/answer | Twilio call answer — returns TwiML greeting |
POST /api/twilio/{slug}/gather | Twilio speech gather — processes caller input |
POST /api/twilio/{slug}/status | Twilio call status callback |
GET /calls | Call logs list with transcripts |
GET /calls/{id} | Individual call detail with full transcript |
GET /calls/coaching | Call coaching dashboard |
GET /settings/ai-persona | AI persona editor |
GET /settings/voice-ab | Voice A/B testing |
GET /settings/conversation-flows | Custom conversation flows builder |
Key Files
| Type | Path |
|---|---|
| Service | app/Services/ConversationEngine.php — core AI handler |
| Service | app/Services/VobizService.php — Vobiz webhook responder |
| Service | app/Services/TwilioService.php — Twilio voice handler |
| Service | app/Services/IntentRouter.php — keyword intent detection |
| Service | app/Services/FailoverService.php — OpenAI failover wrapper |
| Model | app/Models/CallLog.php |
| Model | app/Models/AiActionLog.php |
| Controller | app/Http/Controllers/CallLogsController.php |
| View | resources/views/calls/ |
| Table | call_logs — transcript, sentiment, quality_score, duration, caller_number |
| Table | ai_action_logs — intent_detected, response_generated, failover_activated |
| Table | ai_prompt_versions — versioned AI prompt history |
What's Complete
- Vobiz webhook endpoints — full answer/input/hangup cycle with XML responses
- Twilio webhook endpoints — TwiML answer/gather/status cycle
- OpenAI GPT-4o conversation engine with multi-turn history
- Call session management (conversation state stored across webhook calls)
- Full call transcripts saved to DB on hangup
- Post-call sentiment analysis (positive/neutral/negative)
- AI conversation quality scoring (0–100)
- AI call summary and call outcome detection
- Call coaching dashboard with coaching notes and quality trends
- AI prompt versioning with Super Admin view and revert
- Voice A/B testing (compare two prompt variants)
- Custom conversation flows builder (configurable call paths)
- AI persona editor (name, style, greeting, escalation message)
- Emergency escalation + human handoff queue
- Healthcare demo: live doctor list, slot availability, appointment booking via voice
- Intent detection (9 intents) with keyword matching before OpenAI call
- Failover system with intent-aware fallback responses
- AiActionLog records every significant AI event
What's Deferred
- Twilio hangup wired for call_negative_sentiment trigger (Vobiz is wired; Twilio needs the same)
- Multilingual TTS (Hindi, Tamil voice — needs different TTS provider)
- Accent/local tone engine (needs multilingual voice model)
- Real-time call monitoring (listen in on live calls)