The scenario
Wanderloom coordinates specialist agents through a LangGraph supervisor: a destination researcher gathers options, an itinerary planner sequences days around travel time, a budget agent keeps costs in check, and a reviewer catches clashes before the plan reaches the user. A good first multi-agent project — small enough to finish, rich enough to teach handoffs and shared state.
System architecture
01 Traveller brief Dates, budget, style
02 Supervisor Delegates work
03 Research agent Places + events
04 Planner agent Day-by-day plan
05 Budget + review Costs and clashes
06 Itinerary Shareable plan
High-level flow of Wanderloom, from input to output.
What you will learn
- Supervisor and handoff patterns between agents
- Shared state and structured outputs across a graph
- Constraint checking with a reviewer agent
- Turning agent output into a usable UI
What you will build
- Four-agent planning graph
- Budget and travel-time checks
- Streamlit app with editable preferences
- Exportable itinerary
Build plan
- Define the shared state and agent roles
- Build research and planning agents
- Add budget and reviewer checks
- Ship the Streamlit UI
- Export and polish