Sonder

Sonder learns what YOU actually value in a city — the neighbourhoods, the food, the vibe — and scores every place you've been based on your taste. Then tells you exactly where to go next.

Comment

Inspiration

We love Beli. The idea that a restaurant recommendation should come from someone with the same taste as you, not a stranger's aggregate star rating, completely changed how we discover food. But the moment we close Beli and start planning a trip, we're back to TripAdvisor and "Top 10 Things to Do in Tokyo." We wanted to take exactly what Beli got right and bring it to travel — so Sonder was born.

What it does

Sonder is a personal travel rating app where you score individual attractions within a city — the museums, neighbourhoods, food markets, and general vibe — and it computes a cumulative city score weighted by what actually matters. The more you rate, the better it understands your travel DNA. It then generates a personal taste profile (think "The Slow Wanderer") and recommends exactly where you should go next, with specific reasons tied to your rating history. There's also a social feed so you can see what your friends have been experiencing in real time.

How we built it

We built Sonder as a native iOS app using Swift and SwiftUI. The AI backbone runs on Claude (Anthropic's API), which takes your attraction ratings and returns a weighted city score, a personalized summary, and taste profile recommendations — all in structured JSON. We designed the UI around a three-tab architecture — Feed, Lists, and Profile — with a dark premium aesthetic and full-bleed city photography from Unsplash.

Challenges we ran into

Getting Claude to return consistent, well-structured JSON across different rating inputs took more prompt iteration than expected. We also had to carefully design the weighting system for city scores — neighbourhoods and walkability ended up carrying more signal than landmark ratings, which required some tuning to feel right. Shipping a polished SwiftUI interface in 3 hours meant making tough calls about what to cut.

Accomplishments that we're proud of

The cumulative scoring system actually feels meaningful — when you see Tokyo at 9.1 and Cancun at 6.4, it reflects something real about your preferences, not a crowd consensus. We're also proud of the AI-generated city summaries, which reference your specific ratings in a way that feels genuinely personal rather than generic. And honestly, shipping something that looks this good in 3 hours.

What we learned

Personalization is hard to fake — the moment the AI references your actual ratings in its summary ("you seemed to prefer Kyoto's quieter neighbourhoods over Tokyo's sensory overload"), it clicks in a way that generic recommendations never do. We also learned that the social layer — even mocked — is what makes the app feel alive. Without the feed, it's a list. With it, it's a world.

What's next for Sonder

Real social graph — follow friends, see their ratings, filter recommendations by people with similar taste profiles to yours. A "plan my trip" feature that takes your taste profile and builds a day-by-day itinerary for your next destination. Expanding beyond cities to specific experiences — hikes, restaurants, hotels — so your taste profile gets even sharper over time. And eventually, partnerships with travel brands to surface personalized deals based on your actual travel DNA, not just your search history.

Built With

Share this project:

Updates