Curiosity Compass

What is Curiosity Compass?

Curiosity Compass is a guided exploration assistant designed to help users discover lesser-known, culturally rich, and meaningful destinations. It’s made for explorers and lifelong learners. Users input a location and choose 1–5 curiosity tags (like “traditional crafts” or “wildlife observation”) or create their own. We return the most intriguing opportunities based on their selections.

Inspiration

The inspiration behind Curiosity Compass is rooted in exploration. As a Fellow of The Explorers Club and founder of Women in Exploration, Amanda is driven by a desire to uncover stories and places that are often overlooked, whether near home or across the globe. Most travel platforms surface common or trending recommendations, but rarely emphasize deeper cultural context or local insight.

We designed Curiosity Compass with researchers and curiosity-driven thinkers in mind. We wanted a tool that could surprise, educate, and spark discovery—surfacing experiences traditional searches often miss.

How We Built Curiosity Compass

Tech Stack

The frontend is a React app using TanStack Start for server functions, hosted on Vercel.

Development Model

We embraced a rapid prototyping approach: build fast, test often, and refine based on user experience.

Perplexity Sonar API

We tested all available Sonar models and chose Sonar-pro for its strong balance between quality and response time. While the reasoning models didn’t significantly improve results, their “thinking” blocks were valuable for debugging and refining our prompts.

We learned that querying multiple curiosity tags in a single prompt led to skewed results, often dominated by just one topic. To solve this, we split each tag into a separate API call and individually rated the results. This gave us more balanced, relevant recommendations and ensured diversity in the output.

UI/UX Decisions

We focused on practical design choices for a streamlined experience:

Location input modes: Initially offered three (“I’m already there,” “I know where I’m going,” “Surprise me”), but simplified to two: either choose a location or let the app surprise you.
Travel style filters: Considered options like solo, family, or pets—but prioritized discovery. Practical filters can be layered in later once users find something compelling.

Challenges We Faced

Output formatting: Getting the model to return usable, structured JSON required prompt iteration.
Balancing diversity and relevance: We needed logic that fairly represented multiple curiosity tags without over-favoring one.
Response time: Reasoning models sometimes exceeded the 60-second timeout for server functions on Vercel’s free plan. We optimized for fast responses without compromising quality.

What We Learned

Prompt structure matters: Small changes in wording had a big impact on consistency.
Search-based LLMs behave differently: Sonar relies on real-time web data, so understanding its limits and strengths was key.
Constraints helped shape the product: Vercel’s time limits and Sonar’s behavior pushed us to design smarter, more focused queries.

Accomplishments That We're Proud Of

Built a fully functioning MVP over two weekends.
Already used it ourselves and learned new things through the exploration process.
Created a scalable foundation for future product features and expansion.

What’s Next

This MVP solves a clear problem: helping curious people find fascinating, lesser-known experiences around the world that might otherwise go unnoticed or take hours to uncover. But we see even more potential.

In future versions, users will be able to: