Inspiration
We kept seeing the same problem — founders build something, then have no clue who to sell it to first. Customer discovery usually means scattered notes, gut feelings, and wasted conversations. We wanted to make that process less painful.
What it does
You tell it about your product (or just upload a pitch deck / drop in a GitHub link) and it helps you work out which customer segments to go after first. It separates what you believe from what you've actually validated — so you can see where you're guessing vs where you have real signal.
How we built it
Next.js 16 + TypeScript, Claude via AWS Bedrock for the AI, Tailwind for styling. File uploads (PDFs, decks, docs) get parsed server-side. Auth through GitHub OAuth. Simple JSON file for storage — hackathon-appropriate.
Challenges
- Making the AI conversational without it feeling robotic took a lot of prompt tweaking
- Next.js 16 is very new so docs were thin in places
- Keeping the UI simple when the underlying domain model is quite involved
What we learned
- Bayesian thinking maps really well onto customer discovery
- LLMs tend to overweight enthusiasm — you need explicit guardrails for that
- Good types save you at 3am
Built With
- auth.js-v5
- aws-bedrock-(claude)
- bun
- lowdb
- next.js-16
- node.js
- react-19
- tailwind-css-4
- typescript
- vercel-ai-sdk
Log in or sign up for Devpost to join the conversation.