Group travel planning: import links and text, extract places with AI assistance, vote with pairwise comparisons and Elo-style rankings, merge group preferences, build day itineraries with maps, and run a live execution-style flow. Built for HackDuke: Code for Good 2026 (Interactive Media track).

This README documents credits, third-party tools, what the team built vs. what was generated or imported, and AI usage (product and development) for hackathon integrity requirements.

This is not a reskin of a single off-the-shelf “AI travel chat” product. The weekend implementation centers on:

• Domain logic: trip and activity models, Zustand store, routing, and screen wiring (src/lib/store.ts, src/lib/types.ts, src/app/App.tsx, src/pages/*).
• Pairwise voting: Elo updates, comparison pairing, and group ranking merge (src/lib/ranking.ts).
• Import pipeline: TikTok URL handling, transcript fetch path (ScrapeCreators-backed in dev), and orchestration on ImportPage (src/lib/tiktok-*.ts, src/pages/ImportPage.tsx).
• LLM integration layer: client helpers, fallbacks where applicable, and server-side Anthropic calls with hand-written system prompts (src/lib/llm/*, server/index.ts, server/prompts.ts).
• Maps & logistics: Google Maps–related server utilities (geocoding, static map proxy, travel-time enrichment) (server/google-maps.ts, src/lib/geo.ts, related API routes in server/index.ts).
• Custom Vite plugin: dev proxy for TikTok transcript API (vite-plugin-tiktok-transcript-api.ts).
• Seed / demo data: Tokyo activities and coordinates for offline or fallback demos (src/lib/seed-data.ts).

Visual screens in src/app/components/ (e.g. HomeScreen.tsx, VotingScreen.tsx) began as Figma Make exports: layout, Tailwind classes, and Radix/shadcn-style UI primitives were preserved as the design reference; behavior, data, and navigation were implemented or rewired in src/pages/* and the store.

See also ATTRIBUTIONS.md for the short-form Figma Make attribution block.

Other package.json dependencies (e.g. charts, form helpers) are used as licensed open-source libraries; none replace the core pairwise ranking + group trip product story above.

• Anthropic Claude (via API) is used to:
  • Extract structured activities/places from pasted text or transcripts.
  • Generate or adapt multi-day itineraries from ranked activities and preferences.
• Prompts are defined in server/prompts.ts and maintained by the team; the model returns JSON consumed by our code—we own the schema, validation, and fallbacks in src/lib/llm/ and the Express handlers.

If the API key is missing or invalid, the app is intended to remain demo-capable using seed data and client-side fallbacks where implemented.

Team: edit this subsection before submission with accurate names and scope.

• Tools used Claude for Brainstorming and Cursor with Composer 2 and Opus 4.6
• How we used them: e.g. scaffolding components, refactoring, debugging, drafting prompts, writing tests, documentation.
• What we verified: we reviewed, tested, and integrated all generated code; we did not ship unreviewed copy-paste as final logic.

npm install

Copy .env.example to .env and set:

• ANTHROPIC_API_KEY — Claude on the server (optional for demo if fallbacks/seed suffice).
• GOOGLE_MAPS_API_KEY — maps/geocoding features.
• SCRAPECREATORS_API_KEY — TikTok transcript path in dev (optional).

npm run dev:all

• Vite (UI) is typically at http://localhost:5173 (proxies /api to the backend).
• API default port: 3001 (override with API_PORT if needed).

npm run dev

npm run dev:server

npm run build

Third-party packages remain under their respective licenses. Figma-exported and shadcn-derived UI files are subject to those projects’ terms. Unsplash images require compliance with the Unsplash License.

• Credited: Figma design source, shadcn/ui, Unsplash, npm ecosystem, Anthropic, Google Maps, ScrapeCreators.
• Built by the team: trip/vote/ranking flow, store, routing, server API, prompts, import pipeline, maps enrichment, and integration—not a thin skin on a single generic AI chat template.