🎙️ Voice-Driven Agentic AI Platform
Semantic Routing · Agentic RAG · MCP Client-Server · LangGraph · WebRTC Voice · Azure OpenAI
🔁 LangGraph
🧩 MCP Client-Server
🧠 Azure OpenAI GPT-4o
🎙️ WebRTC + LiveKit
📚 Agentic RAG
🔍 text-emb-3-large
🔊 GPT-4 Preview TTS
🌐 18+ Languages
⚡ FastMCP · Python
VOICE INPUT
→
→
🎚️
LiveKit
Stream
LiveKit
→
🗣️
GPT-4
Preview STT
Azure OpenAI STT
→
SEMANTIC ROUTING
🧭 Semantic Router
text-emb-3-large · Cosine Sim
TRF cache: 14U
>0.75 threshold
→ >0.75→ LangGraph
↓ FALLBACK→ GPT-4o below
⚡ GPT-4o Fallback
Classifies → returns Intent JSON
GPT-4o · Azure OpenAI
↓ FALLBACK route
→
LANGGRAPH ORCHESTRATION
⚙️ LangGraph Orchestrator
Entry · Route · State | LangGraph · Python · Workflow Execution
↙ SOT / Docs path
↘ Tool call path
📚 RAG Agent
Hybrid Search · Agentic RAG
🔍 Vector Search (ANN)
text-emb-3-large
Azure AI Search
📝 BM25 Keyword
Lexical sparse retrieval
↓ ↓ Merge
🔀 RRF Hybrid Fusion
Reciprocal Rank Fusion · re-ranking
↓
🧠 GPT-4o Generation
Augmented prompt · Grounded response
Azure OpenAI GPT-4o
📖 Knowledge Base
• SOPs
• Store Locations
• Service Docs
• Pricing Policy
↓
🔌 MCP Client
JSON-RPC 2.0 request
LangChain MCP Adapter · HTTPX
↓
⚡ MCP Server
FastMCP · Tool Functions
FastMCP · Python
↓ 6 MCP Tools
🏢 Backend Systems
• CARS24 Internal APIs
• Azure Table · Redis
→
📤 Response Node
Format · Optimize
LangGraph · Python
↓
🧑💼 HTL Node
Human Review
[if needed]
LangGraph · interrupt
↓
💾 Memory Node
State Persistence
LangGraph state
→
Azure OpenAI TTS
🔊
GPT-4 Preview TTS
Azure OpenAI TTS API
voice: shimmer
↓
📢
Voice Out
Voice / Voice
Response
<200ms
Voice Latency (WebRTC+LiveKit)
30% Engagement ↑
8% Conversion ↑
Azure OpenAI GPT-4o
text-emb-3-large