Proof: Autonomous Social Proof Engine

Version: 1.0 Demo Status: Hackathon MVP

Overview

Proof is an autonomous AI agent pipeline that transforms customer reviews into high-converting marketing content at scale. The system discovers reviews across the web, obtains customer permission, generates platform-specific video testimonials, and distributes them automatically.

Features

• Automated Review Discovery: Scrapes reviews from G2, Trustpilot, Reddit, and Twitter/X using Rtrvr.ai
• AI-Powered Analysis: Ranks reviews by conversion potential using Cerebras inference
• Permission Management: SMS consent workflow via Twilio
• Multi-Format Video Generation: Creates 3-5 video variations per testimonial using ElevenLabs (voice) and Grok (video)
• Automated Distribution: Posts content to social platforms via Toolhouse.ai
• Real-Time Dashboard: Visual pipeline progress with Server-Sent Events

Architecture

Backend (Node.js/Express)

• Config Module: Environment variable validation
• Logger: Structured Winston logging
• Database: SQLite with better-sqlite3 for pipeline state
• Validation: Zod schemas for type-safe data validation

Agents (Sequential Pipeline)

1. Discovery Agent: Rtrvr.ai integration for review scraping
2. Analysis Agent: Cerebras LLM for review ranking and quote extraction
3. Permission Agent: Twilio SMS for customer consent
4. Generation Agent: ElevenLabs + Grok for video creation
5. Distribution Agent: Toolhouse.ai for social media posting

Frontend (Next.js 15)

• App Router: Modern Next.js architecture
• Dashboard Page: Pipeline control and progress visualization
• Pipeline Component: Real-time stage status display
• VideoCard Component: Video preview and management

API Endpoints

• POST /api/pipeline/start - Start pipeline execution
• GET /api/pipeline/:runId/status - Get pipeline status
• GET /api/pipeline/:runId/sse - Real-time progress stream

Tech Stack

Component	Technology
Backend	Node.js, Express.js
Frontend	Next.js 15, React, TypeScript
Database	SQLite (better-sqlite3)
Validation	Zod
Logging	Winston
Web Scraping	Rtrvr.ai API
AI Inference	Cerebras API
Voice Synthesis	ElevenLabs API
Video Generation	Grok API (xAI)
SMS	Twilio API
Social Posting	Toolhouse.ai API

Prerequisites

• Node.js >= 18.0.0
• npm or yarn
• API keys for: Rtrvr, Cerebras, ElevenLabs, Grok, Twilio, Toolhouse

Installation

1. Clone the repository:

git clone https://github.com/jaffarkeikei/Proof.git
cd Proof

2. Install dependencies:

npm install

3. Set up environment variables:

cp .env.example .env
# Edit .env and add your API keys

4. Create required directories:

mkdir -p storage/videos storage/audio logs

Running Locally

Development Mode (Backend + Frontend)

npm run dev

This starts:

• Backend API server on http://localhost:3001
• Next.js frontend on http://localhost:3000

Backend Only

npm run dev:backend

Frontend Only

npm run dev:frontend

Production Mode

npm run build
npm start
npm run start:frontend

Usage

1. Open http://localhost:3000 in your browser
2. Enter a company name or URL
3. Click "Start Pipeline"
4. Watch real-time progress as agents execute:
   • Discovery: Scraping reviews
   • Analysis: Ranking and extracting quotes
   • Permission: Sending SMS consent requests (optional)
   • Generation: Creating video variations
   • Distribution: Posting to social platforms
5. Preview generated videos in the dashboard
6. Download or share videos as needed

Project Structure

/
├── src/
│   ├── agents/
│   │   ├── discovery.js      # Rtrvr integration
│   │   ├── analysis.js       # Cerebras integration
│   │   ├── permission.js     # Twilio integration
│   │   ├── generation.js     # ElevenLabs + Grok
│   │   └── distribution.js   # Toolhouse integration
│   ├── api/
│   │   └── routes.js         # Express API endpoints
│   ├── pipeline/
│   │   └── orchestrator.js   # Pipeline coordination
│   ├── utils/
│   │   ├── config.js         # Environment config
│   │   ├── logger.js         # Winston logging
│   │   ├── database.js       # SQLite operations
│   │   └── validation.js     # Zod schemas
│   └── server.js             # Express app entry point
├── app/
│   ├── layout.tsx            # Root layout
│   ├── page.tsx              # Dashboard page
│   └── components/
│       ├── Pipeline.tsx      # Pipeline visualizer
│       └── VideoCard.tsx     # Video preview card
├── storage/                  # Generated content
│   ├── videos/
│   └── audio/
├── architecture.json         # PDD architecture definition
├── package.json
├── .env.example
└── README.md

Development Workflow

This project uses Prompt Driven Development (PDD) with Claude Code:

1. Architecture defined in architecture.json
2. Generate prompts: pdd generate
3. Sync code: pdd sync <module_name>
4. Iterate and refine

Performance Goals

• Pipeline execution: < 60 seconds end-to-end
• Review discovery: < 10 seconds
• AI analysis: < 5 seconds (Cerebras fast inference)
• Video generation: < 30 seconds per variation
• Distribution: < 5 seconds per platform

Demo Flow (5 Minutes)

Minute 1: Introduce problem - "Companies have 100s of reviews but only use 1-2"

Minute 2: Enter company name → Show live scraping (Rtrvr)

Minute 3: AI analyzes and generates 3 video variations in 30 seconds

Minute 4: Auto-post one video to LinkedIn (Toolhouse action)

Minute 5: Show analytics + "This ran autonomously while I was talking"

Known Limitations (MVP)

• Single company per run (no batch processing)
• No authentication (demo only)
• SQLite database (not production-ready)
• Local file storage for videos
• Manual approval required for distribution
• Limited error recovery

Future Enhancements

• Multi-company management
• Advanced video editing capabilities
• Payment processing integration
• Cloud storage for media files
• Webhook support for async processing
• Analytics dashboard with engagement metrics
• A/B testing for video variations

Troubleshooting

API Errors

• Verify all API keys in .env
• Check API rate limits and quotas
• Review logs in logs/ directory

Database Issues

• Delete proof.db to reset database
• Check file permissions on storage directories

Video Generation Slow

• Grok API may be rate limited
• Consider reducing video variations (5 → 3)
• Check network connectivity

License

MIT

Contributing

This is a hackathon demo project. For production use, please review security, scalability, and error handling requirements.