Culturo

Inspiration

One of the ways that people going about learning about culture is by learning language. This is partially because that is probably the most streamlined way to do so and there are a plethora of resources for language learning. What we wanted to do is different. We want to give users an opportunity to learn to be a local whether that's by exploring the history, music, food, and landmarks from a country or by engaging in conversation with an agent to learn what its like to be a local.

What it does

WorldWise is an AI cultural immersion companion that delivers personalized snapshots via voice, visual content, and trivia. Voice-first interface — Engage in conversation with a Vapi Voice Agent to learn what it's like to be a local Cultural content feed — movies, news, music, restaurants, and attractions that make a country what it is Real-time trivia — dynamic questions from fetched content; stores memory about their user Agentic orchestration — Spotify, NewsAPI, Reddit, TripAdvisor, Wikipedia, Google Maps etc

How we built it

Frontend (React), Backend (Python/Flask), Multi-agent framework for coordinated workflows Asynchronous API integration with aiohttp, Health, orchestration, content feed, trivia, and voice endpoints Orchestrator routes user queries, ContentFeed agent gathers real-time data, TriviaAgent generates questions from content; Parallel API calls to reduce latency API integrations: Spotify (local music), NewsAPI (current events), Reddit (slang/community trends), TripAdvisor (restaurants, attractions), Deepgram (multilingual STT), Anthropic Claude Haiku (reasoning), Vapi (Voice AI)

Challenges we ran into

• Integrating the sheer amount of APIs was difficult because each had different types of documentation + different hurdles to cross
• deepgram + Vapi was difficult because we were trying to include language support beyond just English; this required implementing a language detection script through deepgram and accordingly prompting vapi to respond in that language to come off as more personable ## Accomplishments that we're proud of
• a working MVP for 3 cultures (Japan, Spanish, France)
• a solid intuitive front end with a content feed, trivia, and voice-first chatbot
• support for english, spanish, french, and japanese + scalable for other languages
• Letta Integration that remembers user progress; creates personalized experience for users; key for cultural immersion ## What we learned
• Technologies: Letta, Vapi, Deepgram, Claude
• While agentic workflows are always subjected to latency, there are ways to perform simultaneous processing to cut it down
• A voice first engagement approach makes it more accessible and personable for the user ## What's next for Culturo
• Coming out with support for more cultures; our system is designed to be scalable (just have to fully build out the integrations)
• reducing latency with processing content and setting up the voice agent
• gathering feedback from users to see if there are other ways to deliver the culturally immersive experience we are passionate about

Built With

• arize
• claude
• deepgram
• flask
• javascript
• letta
• newsapi
• python
• react
• spotify
• tripadvisor
• vapi