MapMyTrip

The problem

Planning a trip or a group outing means jumping between search engines, Google Maps and group chats. Everyone has a different budget and taste, and nobody knows whether a place will be open at the planned time or whether the distances make sense. Generic AI suggestions tend to invent places or build impossible schedules.

The solution

Two flows: Trips builds a day-by-day itinerary from a prompt, and group plans let each participant join through a link without signing up, share their budget and preferences, and vote between 3 plans proposed by the AI. Key decisions:

• Real places, not made-up ones: generation is a LangGraph graph that searches Google Places and feeds the model distances and opening hours. The code then checks the plan (places closed at that time, duplicates, too far apart or over budget), repairs it when possible and computes real travel times with the Routes API.
• Reproducible evals: fixed scenarios graded in code and by a larger model, traced in LangSmith. Prompt or model changes are compared with data, not gut feeling.
• Curated places: beyond Google, an admin panel for hand-picked places, with a partner badge for paying businesses.
• Architecture: Nuxt 4 with domain layers. The Places key never reaches the browser, inputs are validated with Zod, login uses Google OAuth and WhatsApp share images are generated on the server.
• Releases with release-please and Conventional Commits.

The outcome

In beta (v0.4.0). Itinerary generation, group plans with voting and closing, the places admin panel and the evals all work. In progress: a place panel with photos, reviews and hours, two columns on desktop and a bottom sheet on mobile.