Cove

Inspiration

We were inspired by frustrations and aspirations that we have personally witnessed or experienced because we wanted to create a product that tackles real pain points. With that in mind, we choose the themes of financial and travel planning as we all have had frustrations with travel planning as students. We wanted to leverage our benefits as well as our flexibility to explore opportunities in creative tourism.

What it does

Cove is a platform designed for students all over the world, where you can quickly get personalized traveling plans that guarantees competitive low prices via our AI-Agent, Coot. At the same time, you can browse manually and find hidden gems on this platform that are reviewed and verified by Cove users.

How we built it

We built the app using Next.js with TypeScript, hosted on Vercel for smooth deployment. ElevenLabs powers our conversational AI — handling both the voice input and output through an Agent connected to our backend. Gemini 2.5 (via the Gemini API) generates tailored travel recommendations using user data such as budget, interests, and destination. Drizzle ORM with Neon Postgres manages all travel summaries and recommendations in a relational database. Clerk handles user authentication and session management, keeping user data synced across sessions. We used webhooks to connect ElevenLabs’ conversation summaries with our database, ensuring user data (saved recommendations, favorites, etc.) updates automatically after each AI interaction. Throughout the hackathon, we collaborated using GitHub for version control and VS Code Copilot (Claude Sonnet 4.5) for faster coding and debugging assistance.

Challenges we ran into

One of our biggest challenges was setting up webhooks to keep data synchronized between ElevenLabs, Clerk, and our Postgres database. Making sure user authentication and AI-generated summaries matched the right accounts took a lot of testing and debugging. We also faced time constraints connecting the Gemini and ElevenLabs APIs together smoothly — both had excellent documentation, but handling real-time voice data while maintaining state across the conversation wasn’t trivial. Another challenge was budget allocation logic. We wanted the AI to recommend activities that stayed under budget per day, which required designing a clean data model and smart cost calculations that updated dynamically. In terms of collaboration, a challenge that we only realized later was that we didn't align on the styling of the frontend with the backend early enough. This resulted in a disconnected backend - UI system and the frontend UX.

Accomplishments that we're proud of

We were happy that we were able to flush out our idea in the end (by the afternoon) as we spend quite some time on it. We believe in that it is a strong idea and can be iterated on even after the hackathon. Technicality wise, we were proud that we were able to get a working voice AI planner fully integrated with real authentication and database persistence in less than a day. While we couldn't save conversations, we successfully synced AI-generated data with our backend using webhooks. Building a system that feels genuinely useful for students — something we’d actually use ourselves. Pulling off an end-to-end stack: Gemini, ElevenLabs, Clerk, Drizzle ORM, Neon, and Vercel — all working together.

What we learned

In terms of technicality, we learned how to combine multiple APIs and frameworks in a real-world project and handle async data flow between them. Integrating webhooks and auth systems taught us how crucial it is to plan data flow early. We learned the importance of clear documentation and debugging strategy — even small mistakes in webhook URLs or API payloads can cause big delays. Most of all, we learned how to prioritize ruthlessly under a strict timeline and keep the product moving forward, even when things broke.

What's next for Cove

Within the tight time frame, we prioritized on the main functionality of our product which is the AI-Agent that helps our user scrape the internet and find the lowest competitive prices for activities that the user is interested in (using ElevenLab's AI API). The next steps for Cove is to get our first batch of users to review the activities on the platform, and encourage our users to input activities that are not found elsewhere on Booking.com or such. With that, we can help Cove users find hidden gems (i.e. restaurants, activities, workshops, etc.) that are updated and verified within a community of students. Our AI-Agent will continue to assist our users in finding the cheapest and tailored activities, events, and booking for our users as our own database grows.

Built With

• clerk
• elevenlabs
• fastapi
• figma
• javascript
• next.js
• node.js
• python
• react
• tailwind
• typescript