✈️ AirlineNexus Architecture

Multi-Agent AI System - From User Input to Response

INPUT
🖥️ User Interface Layer
👤
User Input
Natural Language Query
Streamlit Web UI
Users interact through a modern web interface built with Streamlit, asking questions in natural language about flights, policies, or support needs.
💻
CLI Interface
Command Line Access
Python CLI
Alternative command-line interface for developers and power users to interact with the system programmatically.
⬇️
ROUTE
🎯 Coordinator & Routing Layer
🎯
Coordinator Agent
Intent Classification & Routing
Strands Framework
Central orchestrator that analyzes user intent, classifies queries, and routes them to the appropriate specialized agent while maintaining conversation context.
🛤️
Request Router
Multi-Agent Orchestration
Context Management
Intelligent routing system that determines the best agent for each query and manages multi-step workflows across different agents.
⬇️
AGENTS
🤖 Specialized Agents Layer (4 Agents)
🛫
Flight Agent
Flight Operations
MCP Server
1
Handles flight search, booking, and management using MCP (Model Context Protocol) server for real-time flight operations and reservations.
📋
Policy Agent
Policy & Rules
TiDB Vector DB
2
Provides policy information using TiDB Serverless Vector Database for semantic search of airline policies, baggage rules, and regulations.
🎧
Support Agent
Customer Support
Python Tools
3
Manages customer support issues using custom Python tools for ticket creation, issue analysis, and escalation management.
💬
General Agent
General Queries
General Tools
4
Fallback agent for general travel information and conversational assistance that doesn't fit into other specialized categories.
⬇️
SERVICES
🔧 External Services & Data Layer
🚀
MCP Server
Flight APIs
FastMCP
Model Context Protocol server providing flight search, booking, and status tools for the Flight Agent with real-time airline data access.
🗄️
TiDB Vector DB
Policy Database
Vector Search
TiDB Serverless Vector Database storing airline policies with semantic search capabilities using embeddings for accurate policy retrieval.
🛠️
Python Tools
Support System
Custom Tools
Custom Python tools for support ticket creation, email integration, issue tracking, and escalation management workflows.
🤗
Embeddings
Semantic Search
Sentence Transformers
Sentence Transformers for generating text embeddings used in policy search and semantic understanding of user queries.
⬇️
MODELS
🧠 AI Models Layer
🥝
Kiwi LLM
Domain Expert
Think, Act, React
Specialized domain model for airline knowledge and support scenarios, providing industry-specific expertise and empathetic customer service responses.

🔄 Complete Flow Process

1

User Input

User asks a question through Streamlit web interface or CLI about flights, policies, or support issues.

2

Intent Analysis

Coordinator Agent analyzes the query, classifies intent, and routes to the appropriate specialized agent.

3

Agent Processing

Specialized agent processes the query using their specific tools: MCP server, TiDB Vector DB, or Python tools.

4

AI Model Response

Kiwi LLM generates intelligent responses based on retrieved data and context understanding.

5

Response Delivery

Final response is formatted and delivered back to the user through the same interface with actionable information.