toolsDetail.backToAllTools
toolsDetail.projectRecordtoolsDetail.categories.mapsTraveltoolsDetail.statuses.openSourcetoolsDetail.updatedOn

ThaiGuide-AI Planner

A Kaggle-built AI travel planner for Thailand — reasoning-based routing, live voice guidance, and grounded, real-time maps.

ThaiGuide-AI Planner turns a conversation into a real Thailand itinerary: it reasons through routes and logistics, talks back through a live voice guide, grounds its answers in Google Search and Google Maps, and plots the whole trip on a real-time map.

toolsDetail.overview

ThaiGuide-AI Planner is an open-source Kaggle travel-planning app built around a single persona, Somsri, a warm and knowledgeable Thai tour guide who stays in character across three modes: a live voice guide, a trip planner, and a text-based local chat. Pick a destination (Bangkok, Chiang Mai, Phuket, Krabi, or Ayutthaya), add interests, and it produces a day plan with real attractions, restaurants, and timing.

Route planning runs through two tiers. By default, itineraries are generated with a fast, grounded model call. Flip on Deep Thinking mode and the same request routes to a reasoning model configured with a large thinking budget, prompted to work through traffic, weather patterns, and cultural timing and explain its logic — the difference between 'here's a plan' and 'here's a plan and why it holds up.'

Every itinerary is grounded rather than invented from the model's own memory: requests attach both Google Search and Google Maps as tools, so place names, hours, and prices are checked against live sources, and the response carries grounding citations back with it. The Live Guide mode uses the same Somsri persona for a real-time, spoken conversation — talk through what kind of trip you want, and the transcript from that conversation feeds straight into the planner to generate the itinerary, no retyping required.

The grounded place references double as map data. Once an itinerary comes back, the app pulls the named locations out of the grounding results and draws an actual driving route between them on a live map, with total distance, duration, and stop count calculated from the routed legs — not just a list of names, but a plotted trip.

toolsDetail.keyFeatures

Reasoning-based route calculation

A default fast path for straightforward itineraries, plus an opt-in Deep Thinking mode that routes complex logistics — traffic, weather, cultural timing — through a reasoning model with a large thinking budget and asks it to show its work.

Google Search and Google Maps grounding

Itinerary requests attach both Search and Maps as tools, so attractions, restaurants, hours, and prices are checked against live sources rather than generated from memory, with grounding citations returned alongside the plan.

Live voice guide

A real-time, bidirectional voice conversation with the Somsri persona over the Gemini Live API. The conversation transcript feeds directly into the trip planner to generate an itinerary from what you said, not what you typed.

Real-time route mapping

Grounded place references are plotted on a live Google Map, with an actual driving route drawn between stops and total distance, duration, and stop count calculated from the routed legs.

Local chat companion

A separate, stateful text chat for etiquette, bargaining, food, and culture questions, staying in the same guide persona and steering off-topic questions back toward Thailand.

Spoken itineraries

Generated plans can be converted to speech, so a finished itinerary can be read back aloud instead of only displayed as text.

toolsDetail.architecture

FrontendA React 19 and TypeScript single-page app built with Vite. Maps render through the Google Maps JavaScript API and its UI Kit web components, loaded directly in the client.
AI orchestrationThe @google/genai SDK coordinates several Gemini models for different jobs: a fast model for grounded itinerary generation, a reasoning model with an extended thinking budget for deep logistics analysis and chat, a native-audio live model for the voice guide, and a text-to-speech model for spoken playback.
Grounding and mapsItinerary calls attach Google Search and Google Maps as tools. The grounding data returned with each response is used to extract real place names, which then drive a Directions Service pass that renders the route and totals distance, duration, and stops on the map.
Audio pipelineMicrophone input is captured through a 16kHz audio context and encoded to PCM16 for the Live API; returned audio is decoded from base64 PCM and played back through a queued buffer pipeline, with careful cleanup to stay under browser audio-context limits.
HostingA client-only app scaffolded via Google AI Studio, with no backend server — Gemini API calls run directly from the browser using an API key.
React 19TypeScriptVite@google/genai (Gemini API)Google Maps JavaScript APIGoogle Search grounding

toolsDetail.bottomCtaQuestion

toolsDetail.browseAllTools