VibePitch

Inspiration

When founders want to practice pitching, they usually have to know someone in their network who understands venture capital and entrepreneurship. However, with Generative AI & LLMs, we can now emulate a VC who can sharpen you from your own home. This is what VibePitch tries to do.

What it does

VibePitch lets founders practice their startup pitch against an AI venture capitalist before the real thing. You upload your pitch deck as a PDF, and the app extracts the content and feeds it to an AI agent that has "read" your deck. Then you jump into a live, real-time voice conversation where the AI VC grills you with tough questions, challenging your market size, unit economics, competition, and defensibility. It's like a mock partner meeting you can run anytime, so you walk into the real room ready for anything.

How we built it

The initial Claude Code prompt for this project was to make a plan to build a platform that lets users submit a deck and pitch to an AI VC with help from the ElevenLabs API. The main goal of the prompt was to interrogate it on tech stack choices (e.g., TypeScript - Next.js + Elysia, Next.js TS-only, Next.js + FastAPI) and about the specificities of the ElevenLabs API itself. Claude did a thorough doc search, and thoughtfully considered how to make all aspects programmatic instead of GUI-based. Finally, I used Hostess to deploy VibePitch.

Challenges we ran into

Some challenges included the misconfiguration of the ElevenLabs model, inability for the deck to be parsed, and the upgrade from 2nd gen model to 3rd gen model. Using Claude Code to debug with logs as well as more thorough doc search, I managed to resolve these errors pretty quickly.

Accomplishments that we're proud of

The AI VC is really good at interrogating the user on their business, not trying to assume anything in particular — the intended outcome.

Built With

• bun
• docker
• elevenlabs
• framer-motion
• hostess
• nextjs
• openai-api
• pdf-parse
• react
• tailwind
• typescript