Creep Slapper 9000

• 

Inspiration

I keep making the same mistake at hackathons: I pitch a whole startup, over-scope, under-deliver, and miss the deadline.

I wanted a tool that grabs me by the collar and says, "Here's what you actually need to ship."

The idea-trimmers out there cut blindly — with no sense of what a specific competition rewards. So, Scope-Creep Slapper 9000 was born: cuts with context, not cuts for the sake of cutting.

What It Does

Paste your wildly over-ambitious idea (type it, dictate it, or load a demo creep), add context — a hackathon link, raw rules, or both — and hit SLAP.

You get a verdict:

• What to keep: The essential features that make your idea work.
• An MVP ship checklist: A practical list of tasks you can copy with one click.
• Your creeps, struck through: Features that got cut, each with a "Why slapped" rationale.

It's deliberately demo-minded: flashy UI and polish survive the cut when they're what make a one-minute demo land, even though a blind trimmer would kill them.

How We Built It

We built a single-page application using Vite, React, and TypeScript, wrapped in a light neo-brutalist design.

The backend uses three Vercel serverless functions to ensure API keys never touch the browser:

• DeepSeek: Makes the cut by evaluating the idea against the competition's requirements.
• Firecrawl: Turns a hackathon link into readable criteria text that feeds into the prompt.
• Deepgram: Streams hold-to-talk audio into the idea input box.

The prompt is the product's brain. It trims approximately 80% of an over-scoped idea based on the competition at hand — its theme, judging criteria, and constraints — instead of trimming in a vacuum.

Challenges We Ran Into

1. Firecrawl returned 401 errors.

The integration returned 401 errors on both API versions, initially making it look like an endpoint issue. However, identical errors pointed to an invalid API key. We tested one variable at a time, confirmed the diagnosis, corrected the key, and got a live 200 response.

2. Deepgram transcribed empty audio.

Our test sample contained empty audio. The problem wasn't our code; it was a dud file. We fixed it by switching to a known-good sample.

3. Vercel's serverless module resolution.

The classic serverless gotcha: ERR_MODULE_NOT_FOUND forced us to inline the slap prompt directly into the function.

Every issue was caught by testing each component before moving on — never by discovering it at demo time.

Accomplishments That We're Proud Of

• A working, end-to-end pipeline: Three external services working together, verified live rather than mocked.
• A satisfying verdict experience: Slap choreography, staggered creep cards, and a rationale for every cut.
• Genuine context awareness: Paste the Build With AI: Basics page, and the cuts respect that competition's requirements.
• A disciplined development process: Six slices, each verified before moving to the next, with zero demo-day surprises.

What We Learned

Plan before you build, and gate every step on runnable evidence — a command that proves the step works, plus a failure case that proves the fallback works.

We learned to:

• Keep secrets server-side by default.
• Distrust a single test sample.
• Isolate one variable at a time when debugging.

That last lesson paid off: diagnosing the 401 errors took minutes, not hours, because we tested systematically instead of guessing.

What's Next for Scope-Creep Slapper 9000?

We've got plenty more slapping to do:

• Slap history and shareable links: Save previous slaps and share results with teammates.
• Deeper criteria parsing: Weight feature cuts against actual judging rubrics.
• Checklist exports: Export the MVP checklist to Markdown or GitHub Issues.
• A public demo URL: Let anyone try a slap without having to run the project locally.

Built With

• deepgram
• vite