Learn2Play.Store

Inspiration

I love playing board games but hate having to explain them to my friends. Learn2Play is a solution to game nights thwarted by complex rules, making it easier than ever for anyone to start playing any board game they own.

What it does

Store: Buy and sell game rules on our marketplace, or download them for free. Powered by Stripe, it’s easy to find almost any game you can imagine and start learning it right away.

Text Extraction: Powered by Gemini’s text-embedding-004, our system downloads, chunks, embeds, and stores rules for users in a Supabase database. Using Gemini Flash 2.5, we retrieve rules and answer context-relevant questions, also providing page sources to support the answers.

Video Interface: With Tavus and 11 Labs, we bring our agent to life through a real-time video interface. Users can talk and interact with it as they navigate the 3D landscape of their game and questions.

Challenges we ran into

Storing and embedding data—and giving our custom Tavus persona access to that data—was a major challenge. We had to create a suite of tests using Bolt.New to verify database chunking and embedding, along with custom edge functions to support agent access.

Using only Bolt.New and public documentation, I was able to troubleshoot and resolve all outstanding issues and build an end-to-end experience I'm proud of.

Accomplishments we're proud of

I'm most proud of our RAG and custom agent implementation. It feels naturally integrated into the product, which also includes a standard web experience for things like the marketplace and account management that will feel intuitive to users.

The ability to quickly change game context and access embedded data sources for fast, accurate retrieval made it a powerful tool for our video agent.

What we learned

I learned a lot about context limits and video agents while working with Tavus. I also used edge functions and embeddings heavily for the first time in this project, specifically with Supabase.

Getting a high level of accuracy required a lot of prompt and context trial-and-error, which could be frustrating at times—especially considering how many agents were involved throughout the project. While the final video agent isn't perfect, it's reached a very proficient level when accessing rules via our Gemini Flash agent and edge functions. I was shocked I could implement such a new tool so effectively using just Bolt.New prompts.

What's next for Learn2Play.Store

Right now, managing rulebooks and multipart questions is still a bit tricky. For games like Warhammer, where more than five PDFs are needed just to play, it can be tough for the system to retrieve and reason through all the parameters of a question. For the basics of most games, the system works great, but more complex tasks will likely require additional functionality.

I'd love to bring more games—both free and paid—into our store to support creatives in the gaming space. Many are selling rule sets and 3D sculpts entirely online, though it's still a niche market. I believe digital downloads and 3D prints represent a new era of board gaming, allowing anyone to quickly create and distribute games—if only people had the time to learn the rules!

Built With

• bolt.new
• elevenlabs
• stripe
• supabase
• tavus