Skip to content
AI Workshop

TripMate · Jag Reehal's AI workshop

Build a Trip-Planning Agent

A two-path workshop in Python with Pydantic AI or TypeScript with the Vercel AI SDK. Pick your language, build the TripMate foundations, then choose your own adventure across the tracks.

This workshop teaches you to build a trip-planning agent called TripMate. First pick your language. The two paths run in parallel and hit the same checkpoints with idiomatic code:

  • Pydantic AI (Python): pydantic-ai/
  • Vercel AI SDK (TypeScript): vercel-ai-sdk/

How the workshop works

Section titled “How the workshop works”

Everyone starts with the Foundations. After that the path forks, and you choose where to go. Do one track, several, or stop after foundations. Jump straight to a track if you already know the basics.

Start here · everyone Foundations f1–f7 · the augmented-LLM primitives
PY Pydantic AI TS Vercel AI SDK
  • Patterns p1–p7 orchestrate & compose Live · in the room
  • RAG r1–r2 ground it in your data Live · in the room
  • Full-Stack the agent behind a UI Live · in the room
  • MCP tools from a server Self-serve · take-home
  • Resilience errors as data Self-serve · take-home

Foundations first, then choose your own adventure. Both language paths share these tracks — pick one above to start.

Foundations (f1–f7): Call an agent, read the loop, get typed output, write a tool, shape which tool the model reaches for, and more. The shared starting point for everyone.

Discussion: the workshop ends with 20 minutes of talking, wherever you got to — was working with the model what you expected, what surprised you, what did you learn. Worth it even if foundations is all you did.

Then choose your own adventure across two tiers of tracks.

Live tracks (do in the room):

  • Patterns (p1–p7): Compose and orchestrate. You hold the control flow first (chaining, routing, parallelization, evaluator-optimizer), then hand it to the model with agentic, delegation, and conversation.
  • RAG (r1–r2): Embed your catalogue, rank by similarity, and ground the model in your own data.
  • Full-Stack (web UI): Put the agent you built behind a real web UI, with tools and tool-call cards. Assumes some frontend comfort and adds no new agent concepts.

Self-serve tracks (take-home, any time):

  • MCP: Pull tools in from a separate server.
  • Resilience: Errors as data.

Each challenge has a start/ file with TODOs and a finish/ reference solution.

What you’ll build

Section titled “What you’ll build”

TripMate plans a weekend trip: it looks up a traveller, checks the weather, finds flights, and books a hotel. Take it through the patterns that orchestrate those tools, or give it a face as a streaming React chat app. Your choice.

Get started

Section titled “Get started”