Travel Planner — CrewAI

Problem

Planning a trip end to end (itinerary and budget together) is a multi-step reasoning task that doesn't fit neatly into a single LLM prompt — it needed a workflow, not just a chatbot.

Solution

A CrewAI-based system where a Travel Planner agent produces a detailed itinerary (destinations, activities, transportation) and a Budget agent consumes that itinerary to consolidate cost estimates across categories, delivering one coherent, organized trip plan.

Architecture

CrewAI orchestrates a sequential process: the Travel Planner agent runs first, and its structured output becomes the input for the Budget agent — so budget calculations are always grounded in the actual generated itinerary rather than estimated independently.

Challenges

• —Designing agent roles and task boundaries so outputs compose cleanly instead of duplicating work
• —Getting consistent, structured output from an LLM agent that a downstream agent can reliably parse
• —Keeping the system provider-agnostic enough to swap the underlying LLM backend

Learnings

• —Practical experience with multi-agent orchestration patterns beyond single-prompt LLM usage
• —How sequential agent dependencies simplify reasoning about a pipeline's correctness
• —Designing for extensibility — the repo is structured as a reusable blueprint for further integrations