2026 · Agentic AI
AgentsVille Trip Planner
A trip-planning agent with validated itineraries, mocked tools and an evaluation-driven revision loop.
- Trip request
- Agent and tools
- Validated itinerary
Portfolio project (mocked APIs, no live booking)GitHub ↗
Problem
Show how an LLM agent can produce structured, testable, self-correcting output rather than a single unchecked prompt.
My role
Author of the repository.
Approach
- Pydantic models validate every LLM output; weather and activity APIs are mocked.
- Deterministic checks plus LLM-as-judge evaluate budget, interests and weather fit.
- A ReAct-style revision agent calls tools and iterates until the plan passes the quality gates.
Results
- No quantitative results published; the value is the tested architecture.
Stack
- Python
- Pydantic v2
- OpenAI-compatible API
- pytest