Fernwayer
AI-Powered Travel Platform for Curated Local Experiences
2026 - ongoing
Travel & Hospitality
AI development, web and mobile
38% faster experience discovery
Average time from search to finding a relevant experience decreased from 8 minutes to 5 minutes after introducing the Claude-powered assistant.
24% higher discovery-to-booking conversion
Users engaging with the AI assistant converted from discovery to booking at 14.9% vs. 12% for conventional search.
59% lower zero-result rate
Conversational search reduced searches returning no suitable experiences from 22% to 9%.
31% of bookings influenced by Fern
The Claude-powered assistant contributed to 31% of experience bookings after launch.
About
Project Idea
The client came to Yellow with just an idea and a roadmap. We started from the ground up, first building an iOS app, then expanding into a complete ecosystem: website, admin dashboard, booking system, and an AI travel assistant.
Yellow deployed a Claude Sonnet-powered travel assistant that lets travelers discover relevant, bookable experiences through natural-language search across a catalog of 800+ experiences in 60+ cities across 14 countries.
The Сlient Had
- Product vision
- Initial design
- Feature roadmap
We were responsible for
- Product architecture
- Front and backend development
- AI integration
- iOS development
Team
Project manager
Backend developer
Frontend engineer
AI engineer
iOS engineer
DevOps specialist
Overview
Fernwayer connects travelers with unique, bookable experiences from local experts, using AI-powered discovery to turn natural-language travel requests into relevant recommendations.
For Experience Makers
Experience Makers can set up detailed listings with everything travelers need to know, like descriptions, locations, prices, rules, qualifications, you name it. Every listing goes through an approval process, and Experience Makers can update their offers anytime.
Booking & Availability Management
Providers can hook up external calendars or use Fernwayer’s scheduling to create one-off or recurring events, set limits, and adjust prices throughout the year. Real-time sync keeps schedules up to date and practically eliminates double-bookings. The app’s functionality allows you to define if the event is single-time or a part of a scheduled series, highlight the date and price. You can edit the price according to the price policy of your experience. For example, winter experiences can have a higher price during the Christmas holidays.
For Independent Travelers
Instead of navigating generic travel listings, travelers on Fernwayer can browse and book one-of-a-kind activities led by local experts, tailored to interests, travel styles, and special occasions. Travelers can save activities, share their finds, comment on listings, and build their own collections of places and experiences they want to try. It’s part marketplace, part social network, so users can discover new adventures through both the lists and the community.
Stories
Stories in Fernwayer are posts that allow travelers to show their adventures to other users so they can like or comment on them. It’s possible to add a photo, indicate the uniqueness, and write a small text review about an experience.
Community
Swagger score
Goals
Dreamtrips
AI Assistant
One of the biggest game-changers on the platform is the AI-powered travel assistant.
Most travel platforms stick to simple filters like destination or price. Fernwayer’s AI assistant invites users to naturally describe what they want: "I'm looking for a quiet seaside spot for a weekend with my kids." "Show me unique experiences with restaurants in Barcelona for a small group." "I want an off-the-beaten-path cultural experience during a business trip." The assistant understands the user’s intent and finds relevant, bookable options directly from Fernwayer’s catalog. The magic behind this is a Retrieval-Augmented Generation (RAG) system powered by vector search and function calling. Instead of generating ideas from model knowledge alone, the assistant pulls from Fernwayer’s live inventory through the API, so recommendations are always based on real, available experiences.
Why Claude Sonnet?
Travel requests can be tricky to interpret. They’re often open-ended and need strong context. Claude Sonnet bridges the gap and translates travelers’ natural-language requests into relevant options pulled directly from the platform. It connects the dots, working hand-in-hand with the RAG solution and function calls, so what users get always matches what’s on offer. Combined with Yellow's RAG architecture, evaluation framework, and hallucination guardrails, Claude Sonnet transformed Fernwayer's discovery experience from rigid filtering into grounded conversational search over live, bookable inventory.
Challenges and Solutions
Here are the most notable challenges our team faced while developing Fernwayer.
Non-Standard Experience Inventory
Problem:
Unlike hotels or flights, every Fernwayer experience has unique booking rules, pricing models, availability constraints, and participation formats. Traditional travel marketplace data structures could not accommodate this level of variability.
Solution:
We designed a flexible data model capable of supporting hundreds of unique experience types while maintaining a consistent booking workflow for travelers.
Real-Time Availability Synchronization
Problem:
Because many activities have limited capacity and custom scheduling rules, availability changes had to be reflected immediately to avoid double bookings.
Solution:
We built a real-time calendar synchronization system supported by PubNub notifications, so Experience Makers can manage schedules across multiple channels while keeping inventory accurate.
Other Challenges
Multi-Party Payments
Problem:
The platform needed to distribute payments between Fernwayer and hundreds of independent Experience Makers operating across multiple countries and currencies.
Solution:
We implemented a Stripe-based payout infrastructure that automates commission calculations, supports international transactions, and simplifies financial operations for both the platform and providers.
AI Recommendation Accuracy
Problem:
The LLM could generate recommendations that sound plausible but don’t actually exist in a platform's inventory.
Solution:
We implemented a RAG architecture combined with function calling connected directly to the Fernwayer catalog. The AI retrieves available experiences from the live platform before generating responses, ensuring recommendations remain grounded in real inventory.
Affiliate Attribution Logic
Problem:
A single booking could be influenced by multiple referral sources, making commission attribution difficult.
Solution:
We kept the AI grounded with a RAG setup that ties responses directly to real-time Fernwayer data. The assistant pulls live options before answering, so recommendations actually exist on the platform.
Technology Stack
What technologies we used:
Results
Fernwayer and Yellow deployed a Claude Sonnet-powered travel assistant that uses RAG and live inventory to turn natural-language travel requests into bookable experiences, resulting in 38% faster experience discovery and a 24% increase in discovery-to-booking conversion.
