8000 GitHub - blah-mad/signalcut: AI-search visibility gaps into source-backed Hera video campaigns · GitHub
[go: up one dir, main page]

Skip to content
blah-madPublic

About

AI-search visibility gaps into source-backed Hera video campaigns

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

SignalCut

SignalCut turns AI-search visibility gaps into source-backed video campaigns.

Deployment

Live production app: https://signalcut.pages.dev

Built at Big Berlin Hack for teams that need to turn distribution insight into something they can ship immediately.

What It Does

Early-stage brands increasingly compete inside AI answers, not only on search result pages. A brand can be the best fit for a specific buyer and still lose because answer engines keep recommending larger incumbents.

SignalCut takes visibility context from Peec AI, finds the highest-value prompt gap, pulls source evidence, writes a grounded creative strategy, and generates a Hera video asset.

Example demo scenario:

  • Brand: Attio
  • Competitors: Salesforce and HubSpot
  • Prompt gap: "best CRM for agencies with custom onboarding workflows"
  • Output: ranked opportunity, source evidence, campaign brief, video variants, Hera prompt, completed MP4, and exportable campaign pack

Hackathon Partner Technologies

SignalCut uses at least three Big Berlin Hack partner technologies:

  • Peec AI: visibility context and prompt-gap input. SignalCut normalizes Peec visibility reports into campaign opportunities.
  • Hera: AI video generation. SignalCut sends an opinionated creative prompt to Hera and displays the completed MP4 in the app.
  • Google DeepMind / Gemini: campaign reasoning, opportunity normalization, creative strategy, scripts, and Hera prompt generation.
  • Tavily: live source evidence retrieval for proof-backed campaign claims.

Product Flow

  1. Enter a brand, website, competitors, target audience, and visibility context.
  2. Normalize the visibility context into ranked AI-search opportunities.
  3. Select the strongest prompt gap.
  4. Pull source evidence with Tavily.
  5. Generate the campaign brief and video variants with Gemini.
  6. Create a Hera video job from the selected variant.
  7. Review the completed video and export the campaign pack as Markdown or JSON.

Architecture

SignalCut is a Vite React app deployed to Cloudflare Pages with Pages Functions for provider integrations.

React UI
  -> /api/opportunities  -> Gemini
  -> /api/evidence       -> Tavily
  -> /api/campaign       -> Gemini
  -> /api/hera/jobs      -> Hera
  -> /api/hera/status    -> Hera

Provider keys are never exposed to the browser. They are stored as Cloudflare Pages secrets and used only inside server-side functions.

Tech Stack

  • React 19
  • TypeScript
  • Vite
  • Cloudflare Pages and Pages Functions
  • Gemini API
  • Tavily API
  • Hera API
  • Lucide icons

Repository Structure

src/
  App.tsx              Main product UI and flow orchestration
  lib/api.ts           Browser-safe API client
  lib/scoring.ts       Opportunity scoring
  lib/export.ts        Markdown/JSON campaign export helpers
functions/api/
  opportunities.ts     Peec visibility context -> ranked opportunities
  evidence.ts          Tavily source retrieval
  campaign.ts          Gemini campaign strategy generation
  hera/jobs.ts         Hera video job creation
  hera/status.ts       Hera polling/status endpoint
docs/hackathon/
  signalcut-demo-plan.md
  ai-visibility-video-agent-prd.md
  ai-visibility-video-agent-feature-spec.md

Local Setup

Prerequisites:

  • Node.js 20+
  • npm
  • API keys for Tavily, Gemini, and Hera

Install dependencies:

npm install

Create local secrets:

cp .env.example .dev.vars

Fill .dev.vars:

TAVILY_API_KEY=...
GEMINI_API_KEY=...
GEMINI_MODEL=gemini-flash-latest
HERA_API_KEY=...
HERA_STYLE_ID=

Run locally:

npm run dev

For local Pages Functions parity, use Wrangler:

npx wrangler pages dev dist

Most development work can be done with Vite. Use Wrangler when validating Cloudflare Pages Functions behavior.

Build

npm run build

This runs TypeScript checks for both the Vite app and Cloudflare Functions, then builds the static frontend into dist/.

Cloudflare Deployment

The project is configured for Cloudflare Pages in wrangler.toml.

name = "signalcut"
pages_build_output_dir = "dist"
compatibility_date = "2026-04-25"

Set production secrets:

npx wrangler pages secret bulk .dev.vars --project-name signalcut

Deploy:

npm run build
npx wrangler pages deploy dist --project-name signalcut --branch main --commit-dirty=true

Production URL:

https://signalcut.pages.dev

Security Notes

  • Do not commit .dev.vars, .env.local, .wrangler/, dist/, node_modules/, or artifacts/.
  • Provider keys are server-side only.
  • The browser receives only generated campaign data and provider job status.
  • The app avoids unsupported competitor claims by grounding generated strategy in retrieved evidence.

Scripts

npm run dev        # Start Vite development server
npm run build      # Typecheck and build
npm run preview    # Preview built app locally
npm run typecheck  # Typecheck app and functions

Submission Checklist

  • Live product deployed: https://signalcut.pages.dev
  • Public repository with setup instructions
  • README documents partner technologies
  • README documents APIs and deployment
  • Source code avoids committed secrets

About

AI-search visibility gaps into source-backed Hera video campaigns

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages

0