AirlineNexus

Frontend [Dark Mode]
Frontend [Light Mode]
User looking for flight
User random conversation processing
User random conversation response
Streamlit Server
MCP Server

TIDB CLOUD EMAIL: krupeshpatel7898@gmail.com

Inspiration

The inspiration for AirlineNexus came from our frustrating experiences with traditional customer service not Airline but general like Insurance Company, Online Shopping etc.

when we discovered that different types of customer queries require different expertise and tools - flight operations need real-time data access, policy questions require semantic search through vast documentation, and support issues need structured workflow management. This led us to envision a multi-agent AI system where specialized agents could work together seamlessly, just like how airlines coordinate different departments behind the scenes.

What it does

AirlineNexus is an intelligent, multi-agent AI system that revolutionizes airline customer service by providing a single interface for all travel-related needs. At its core, the system employs a Coordinator Agent that uses advanced intent classification to route customer queries to one of four specialized agents:

The Flight Agent handles all flight operations - from searching available flights across multiple routes and booking management to real-time status updates and schedule modifications. It connects to our MCP (Model Context Protocol) server that provides comprehensive flight tools and APIs.

The Policy Agent serves as an intelligent knowledge base, answering questions about airline policies, baggage regulations, cancellation rules, and travel requirements. It uses TiDB Serverless Vector Database with semantic search capabilities to find relevant policies and provide contextual answers.

The Support Agent manages complex customer issues by analyzing problem complexity, automatically creating support tickets with appropriate priority levels, and handling escalation workflows.

The General Agent acts as a conversational assistant for general travel information, airport details, and miscellaneous queries that don't fit into the specialized categories.

How we built it

We architected AirlineNexus using a sophisticated multi-agent framework built on AWS Strands Agents for orchestration. The foundation starts with our Coordinator Agent, powered by advanced LLM technology, that performs intent classification and intelligent routing to specialized agents.

For the Flight Operations, we implemented a Model Context Protocol (MCP) server using FastMCP, providing standardized tool interfaces for flight search, booking management, and status tracking. This creates a scalable foundation for real airline API integration.

The Policy System leverages TiDB Serverless Vector Database combined with Sentence Transformers for embedding generation. We designed a semantic search pipeline that converts policy documents into vector embeddings, enabling contextual retrieval of relevant airline policies and regulations.

Support Infrastructure utilizes custom Python tools for automated ticket creation, email integration via SMTP protocols, and workflow management. We implemented complexity analysis algorithms that automatically determine issue priority and routing requirements.

The AI Brain runs on Kiwi LLM

The User Interface was crafted in Streamlit with custom CSS.

Challenges we ran into

MCP Server Integration Complexity
Vector Database Performance Optimization
Multi-Agent Context Management

Accomplishments that we're proud of

Successful Multi-Agent Orchestration
Innovative Technology Integration
Scalable Architecture Design
Real-world Application Potential

What we learned

We gained deep insights into the architectural challenges of building truly collaborative AI agents. Managing state, context, and handoffs between specialized agents requires careful design and robust error handling that we initially underestimated. Building production-ready AI systems requires mastering integration between multiple technologies, each with their own quirks, limitations, and optimal use cases. The MCP protocol integration alone taught us the importance of understanding underlying architectures.