TravelBuddy — Multi Agent AI Travel Assistant

A multi agent AI travel planning system that routes user requests to specialized agents for efficient, low latency, and cost aware task execution.

Problem

General-purpose LLM assistants often overuse reasoning and incur unnecessary latency and cost for travel-related tasks that can be handled deterministically via APIs or specialized tools.

Solution

TravelBuddy uses a supervisor agent architecture to interpret user intent and delegate requests to specialized agents for route planning, EV charging discovery, and nearby attraction recommendations. The system dynamically chooses between LLM reasoning and direct API execution based on task complexity.

System Architecture

UserStreamlit UIOrchestratorIntent Analysis & Task RoutingCost & Latency PolicyPrefer deterministic tools • Use LLM only when neededRoute Planning AgentMaps / Directions APIEV Charging AgentCharging Stations & PricingAttractions AgentFoursquare / Places APIResponse AggregationCombine tool outputs • Format responseResponse delivered to user

TravelBuddy routes requests through an orchestrator and a cost/latency policy layer, delegating to specialized agents and aggregating tool results into a final response.

Key Design Decisions

  • Introduce a supervisor agent to centralize intent interpretation and task routing.
  • Avoid unnecessary LLM calls by using tool-specific agents and direct API queries.
  • Design agents as modular components to allow easy extension without modifying the supervisor.

Outcomes

Reduced response latency and LLM usage by routing simple queries directly to tools and APIs. Demonstrated a scalable, cost-aware agent orchestration pattern suitable for real-world travel planning systems.

View implementation on GitHub →