The challenge
Planning a trip usually means stitching together research, flight checks, stay comparisons, and day-by-day sequencing across a dozen tabs. Generic AI chat tools can draft an itinerary, but they often invent places and prices and leave the traveler to verify everything by hand. The problem was turning a loosely described trip idea into a concrete, trustworthy plan without the manual assembly, while keeping the pricing real and the places verifiable.
What we built
Atlazi centers the experience on a conversational planner. A message runs through an AI tool-loop that returns a structured, editable day-by-day itinerary rather than free text. To keep suggestions grounded, the itinerary is built against a real places catalog imported from the Foursquare Open Source Places dataset and stored in a PostGIS-backed database, so the app can find what is genuinely near each stop. Flights, stays, and activities carry live pricing handed off to partner booking sites with no markup, and confirmations, tickets, and delays are tracked in one place afterward.
- Conversational AI planner producing structured, editable itineraries
- PostGIS places catalog from Foursquare Open Source Places for proximity search
- Provider-agnostic AI layer configurable from admin without redeploy
- Single Next.js project serving the marketing site and the app by host
Outcome
The AI layer sits behind a single adapter seam: Anthropic Claude by default, with an OpenAI-compatible endpoint option, configured from an admin panel without a redeploy. One Next.js project serves both the marketing site and the app, split by host in middleware, and maps are rendered client-side with MapLibre GL and PMTiles vector tiles.