DueTube

What Inspired Us Our inspiration came from a shared feeling of "digital inertia." We spend hours watching incredible YouTube videos about everything from ancient history and urban exploration to niche hobbies and scientific wonders. Yet, this digital inspiration rarely translates into real-world action. The spark fades, and we move on to the next video. We wanted to build a bridge between the digital world of content consumption and the physical world of tangible experiences, to make it effortless to act on the curiosity and passion that these videos ignite.

How We Built It DueTube is an AI-native application orchestrated using the CrewAI framework to manage a team of specialized AI agents built on a large language model.

The Scribe Agent: A Python script monitors the YouTube API. When a new video is added, it transcribes the audio and packages the text with video metadata.

The Creative Agent: This agent analyzes the text and brainstorms the core theme for a real-world experience, ideating creative and engaging local activities.

The Research Agent: This agent uses the EXA API for deep, contextual search and the Google Maps Platform API to find relevant and interesting real-world locations.

The Logistics Agent: This agent weaves the research into a compelling narrative and uses the Google Calendar API to find an available time slot and create the event.

The Producer Agent: The final narrative is converted into an audio guide using a Text-to-Speech (TTS) API, and the link is placed in the calendar event.

Challenges We Faced Our biggest challenge was prompt engineering for the Creative Agent. It's easy for an AI to suggest generic activities, but training it to consistently generate high-quality, safe, and genuinely interesting experiences that capture the "spirit" of the source video required significant iteration. Ensuring the handoff between agents was seamless—so the Research Agent perfectly understood the Creative Agent's intent—was also a complex integration task. Finally, managing API rate limits and the asynchronous nature of scheduling across multiple services required robust error handling.

What We Learned This project reinforced our belief in the power of multi-agent AI systems. By giving each agent a highly specialized role (a "Creative," a "Researcher," a "Scheduler"), we could achieve a level of complexity and quality that would be impossible with a single, monolithic prompt. We also learned how crucial structured, narrative-driven output is. DueTube doesn't just give you a place to go; it gives you a reason to go by wrapping the experience in a story. Finally, we gained a deep appreciation for the complexities of integrating multiple, distinct APIs into a single, cohesive workflow.

Built With

• a2a
• crewai
• exa
• gemini
• google-cloud
• mcp
• weave