Inspiration
Our inspiration for IdeaSpark AI came directly from the spirit of this hackathon and the very judge we aimed to impress. We noticed a recurring theme in Greg Isenberg's tweets: an incredible passion for validating startup ideas and a belief that we're in an "AI Gold Rush." He consistently encourages founders to build and ship, but one of the biggest hurdles is knowing if an idea is worth pursuing. We wanted to solve this problem.
The "World's Largest Hackathon" is built on the premise that anyone with an idea can now build it. We were inspired to create a tool that embodies this principle—a first step for any aspiring creator or entrepreneur. IdeaSpark AI is designed to be the friendly, intelligent mentor that gives you the confidence to take your vision from a simple thought to a validated concept.
What we learned
This hackathon was a huge learning experience. Our biggest takeaway was the sheer power and speed of "vibe coding" with Bolt.new. Going from a descriptive prompt to a functional application in minutes is game-changing. We also dove deep into the world of conversational AI. Integrating the Tavus API taught us how to bring a more human, engaging element to data presentation, moving beyond simple text reports to a face-to-face AI interaction. Finally, we learned to think critically about what makes feedback genuinely useful and how to structure AI-generated content to be encouraging yet realistic.
How we built your project
- Foundation with Bolt.new: We started by giving Bolt.new a detailed, zero-shot prompt describing our vision for a clean, intuitive startup validator. Bolt generated the foundational React.js application, including the UI for input and the structure for the results page.
- Core AI Logic: We then defined the "backend" logic conceptually, outlining the different analysis points we wanted the AI to cover (Market, Competition, SWOT, etc.).
- Challenge Integration (Tavus): For the "Conversational AI Video Challenge," we integrated the Tavus API. We set up a workflow where the final validation summary is sent to Tavus to generate a video response from a pre-defined AI mentor replica.
- API Call & Display: We built the client-side logic to call the Tavus API
POST /v2/conversations, retrieve theconversation_url, and embed it seamlessly into our results page using an iframe. - Refinement: We spent the rest of our time refining the user experience, styling the components, and ensuring the final output was polished and professional.
Challenges we faced
Our main challenge was designing the AI's output to be truly valuable. It's easy for AI to give generic advice. We spent significant time structuring our prompts and the final report to ensure the feedback felt personalized and provided concrete, actionable next steps. Another challenge was the time constraint; learning the nuances of both the Bolt.new platform and the Tavus API in such a short period required intense focus and teamwork.
Built With
- langchain
- react
- tavus
- vite
Log in or sign up for Devpost to join the conversation.