Hi Product Hunt!

I'm Atharva, a student founder from Vadodara, India, and Bedrock is my first launch here.

Why I built it

I use AI coding tools every day, and almost every time the first result was broken. The login didn't work, the database forgot things, a page did nothing. The AI wasn't the problem. My instructions were. "Build me a task manager" leaves out dozens of decisions, and the model fills each gap with a guess.

What Bedrock does

It does the architect's job before your AI starts building:

- Asks you a few smart scoping questions (database, sign-in, what happens when something fails)

- Writes a complete spec: pinned versions, a "never do this" list, data structures and a definition of done

- Lets you test it on two AI models side by side

- Lets you refine one section at a time

- Exports it in the format your tool reads (CLAUDE.md, .cursorrules, .windsurfrules and more)

It runs on web, Windows and Mac. The desktop app is bring-your-own-key, and the code is open source:ttps://github.com/Bedrockxai/Bedrock

Who it's for

Students, freelancers, hackathon teams and anyone who has stared at broken AI-generated code and thought "that's not what I meant."

Feedback that would help me most

1. Where do your AI coding projects go wrong today?

2. Do the scoping questions feel useful, or too slow?

3. What would you want in an exported spec that isn't there yet?

I'm building this solo, so every comment goes straight into the next release. I'll reply to all of them. Thank you for trying it!