Inspiration

Planning group trips is chaotic. Between endless group chats, spreadsheets, and browser tabs, friends spend more time coordinating than actually getting excited about their adventure. We asked: what if planning a trip felt as natural as talking to a friend? What if the map updated itself as you brainstormed destinations aloud?

What it does

Nomad is a voice-first collaborative trip planning platform. Join a video call with friends, and our AI travel concierge listens to your conversation in real-time. Say "I want to visit San Francisco" and watch as the 3D globe spins to your destination. Ask for restaurant recommendations and see markers appear on the map with ratings, prices, and reviews. Build your itinerary by speaking naturally—"add that sushi place to our list"—and when you're ready, book everything with a single Solana transaction.

Key Features:

  • 🎙️ Voice-controlled AI travel agent that understands natural conversation
  • 🗺️ Real-time 3D Mapbox visualization with automatic route planning
  • 👥 Multi-user video conferencing so friends can plan together
  • 🍽️ Smart search for restaurants, hotels, and activities with cost estimates
  • 📋 Drag-and-drop itinerary builder that syncs across all participants
  • 💳 Seamless Solana payments for instant booking

How we built it

Voice Pipeline: LiveKit handles real-time audio streaming between users and our agent. Deepgram's Nova-2 model transcribes speech with incredible accuracy, while their Aura TTS gives our agent a natural voice.

AI Brain: Anthropic's Claude powers the agent's reasoning. We implemented custom function tools for searching Yelp, calculating routes via Mapbox Directions API, and managing itineraries—all orchestrated through the Model Context Protocol (MCP).

Frontend: Next.js 14 with TypeScript delivers a responsive UI. Mapbox GL JS renders a stunning 3D globe with terrain, buildings, and smooth fly-to animations. We built custom marker components with rich popups showing Yelp data.

Payments: Solana Web3.js integrates with Phantom wallet for devnet transactions. The agent can request payment confirmation verbally, and users approve with a single wallet click.

Challenges we ran into

Audio synchronization was our biggest hurdle. Browser autoplay policies blocked agent speech until we implemented direct track subscription handlers that bypass React's rendering cycle.

Duplicate agents plagued our early testing—LiveKit would dispatch multiple agents to the same room. We solved this with a single-agent policy check on room join.

Cost estimation required creative thinking. Yelp doesn't provide prices, so we built a heuristic system that estimates costs based on price tier ($-$$$$), location multipliers, and venue type.

Accomplishments that we're proud of

  • The "wow moment" when the globe spins from your current location to your dream destination as you speak
  • Achieving sub-second latency from speech to map update
  • Building a complete booking flow from voice command to blockchain transaction
  • Creating an agent that genuinely feels like a helpful travel companion, not a robotic assistant

What we learned

  • LiveKit's agent framework is incredibly powerful for building voice-first applications
  • Mapbox Standard style with 3D terrain creates stunning visualizations with minimal code
  • The Model Context Protocol (MCP) provides a clean abstraction for giving LLMs access to external tools
  • Solana's speed makes crypto payments feel as instant as traditional ones

What's next for Nomad

  • Flight integration: Search and book flights directly through voice
  • Calendar sync: Automatically block off travel dates and send reminders
  • AI memory: Remember user preferences across trips ("you loved that ramen place in Tokyo")
  • Budget tracking: Real-time spending alerts and group cost splitting
  • Mobile app: Native iOS/Android for on-the-go itinerary access
  • Mainnet launch: Move from Solana devnet to production payments

Built With

Share this project:

Updates