roastify - DISS TRACK GENERATOR

Any2Any Content Transformation Engine - 5 Hour Hackathon Build

🎯 Project Overview

Transform any social media profile (LinkedIn/Twitter) into a personalized rap diss track with AI-generated lyrics, voice synthesis, and auto-generated music video. This showcases our any2any content transformation engine in the most memorable way possible.

🔥 Demo: video https://www.youtube.com/watch?v=xKqb1YUkMPE

Generated rap roast video with AI lyrics, custom beats, and dynamic visuals

🚀 Quick Start

# Clone repo
git clone https://github.com/genmini-ai/roastify
cd diss-track-generator

# Setup environment
cp .env.example .env
# Add your API keys:
# - OPENAI_API_KEY (GPT-5/GPT-4)
# - ANTHROPIC_API_KEY (Claude for backup)
# - ELEVENLABS_API_KEY (Voice synthesis)
# - REPLICATE_API_KEY (Video generation)
# - BROWSERLESS_API_KEY (Web scraping)

# Install dependencies
pip install -r backend/requirements.txt
cd frontend && npm install && cd ..

# Run everything
docker-compose up

# Or run separately
python backend/app.py
cd frontend && npm start

Visit http://localhost:3000

🏗️ Architecture

Input (LinkedIn/Twitter URL)
    ↓
Profile Scraper (Playwright/Browserless)
    ↓
LLM Analysis (GPT-5)
    ↓
Lyrics Generation (GPT-5 with prompts)
    ↓
Parallel Processing:
    ├── Audio Generation (ElevenLabs)
    └── Beat Selection (Pre-made)
    ↓
Audio Mixing (Pydub)
    ↓
Video Generation (Replicate/Local)
    ↓
Multi-format Export (Showcase any2any)

📁 Project Structure

diss-track-generator/
├── README.md                     # This file
├── docker-compose.yml           # One-click setup
├── .env.example                 # Environment template
│
├── backend/
│   ├── requirements.txt        # Python dependencies
│   ├── app.py                  # FastAPI server
│   ├── scraper.py             # Profile scraping (LinkedIn/Twitter)
│   ├── analyzer.py            # LLM-based profile analysis
│   ├── generator.py           # Lyrics generation pipeline
│   ├── audio.py               # Voice synthesis & mixing
│   ├── video.py               # Video generation
│   ├── transformer.py         # Any2any transformation engine
│   └── demo.py                # Demo optimizations
│
├── frontend/
│   ├── package.json           # Node dependencies
│   ├── src/
│   │   ├── App.jsx           # Main React app
│   │   ├── components/
│   │   │   ├── URLInput.jsx  # URL input component
│   │   │   ├── Generator.jsx # Generation progress
│   │   │   ├── Player.jsx    # Audio/video player
│   │   │   └── Transformer.jsx # Format transformer
│   │   └── utils/
│   │       ├── api.js        # Backend API calls
│   │       └── demo.js       # Demo helpers
│   └── public/
│       └── index.html
│
├── prompts/
│   ├── profile_analysis.md   # Profile analysis prompt
│   ├── lyrics_generation.md  # Rap lyrics prompt
│   ├── roast_angles.md      # Roasting strategies
│   └── transformations.md    # Any2any prompts
│
├── assets/
│   ├── beats/                # Pre-made instrumentals
│   │   ├── trap_beat.mp3
│   │   ├── boom_bap.mp3
│   │   └── lofi_beat.mp3
│   └── templates/            # Video templates
│       └── music_video.json
│
├── cache/                    # Demo cache
│   ├── profiles/            # Pre-scraped profiles
│   └── outputs/             # Pre-generated content
│
└── scripts/
    ├── setup.sh             # Environment setup
    ├── warmup.py           # API pre-warming
    └── demo_flow.py        # Demo automation

🔧 Core Components

1. Profile Scraper (scraper.py)

• Handles LinkedIn and Twitter URLs
• Uses Playwright for local dev
• Browserless API for production
• Extracts: bio, posts, work history, education

2. LLM Analyzer (analyzer.py)

• GPT-5/Claude analyzes scraped content
• Extracts: personality traits, buzzwords, achievements
• Identifies roast angles
• Generates personality model for content

3. Lyrics Generator (generator.py)

• GPT-5 writes personalized rap lyrics
• Follows hip-hop structure (intro/verse/hook/outro)
• Maintains rhyme schemes
• Incorporates profile-specific references

4. Audio Pipeline (audio.py)

• ElevenLabs for rap voice synthesis
• Pre-made beats for different moods
• Pydub for mixing vocals + beat
• Real-time streaming for speed

5. Video Generator (video.py)

• Simple but effective visualizations
• Lyrics sync with audio
• Profile images with effects
• Export in multiple formats

6. Any2Any Transformer (transformer.py)

• Showcases our core technology
• Diss track → LinkedIn recommendation
• Diss track → PowerPoint presentation
• Diss track → Academic paper

💨 Speed Optimizations

Critical Path (Must Work)

1. URL → Profile text extraction
2. LLM analysis → Roast angles
3. Generate lyrics
4. Basic TTS audio
5. Display results

Parallel Processing

async def generate_content():
    # Run simultaneously
    tasks = [
        analyze_profile(),     # 2-3s
        generate_lyrics(),     # 2-3s
        prepare_beat(),        # 0.1s (pre-loaded)
    ]
    results = await asyncio.gather(*tasks)

Caching Strategy

• Cache scraped profiles (avoid re-scraping)
• Cache LLM responses (save API calls)
• Pre-load beats and assets
• Use CDN for static files

🎮 Demo Scripts

YC Demo Focus

# Emphasize: Growth, virality, clear monetization
- Quick generation (show speed)
- Share functionality (viral potential)  
- Creator economy angle ($$$)

Afore Demo Focus

# Emphasize: Technical depth, infrastructure
- Show LLM orchestration
- Explain any2any architecture
- Discuss scaling challenges

🚨 Fallback Plans

1. If scraping fails: Manual text input
2. If GPT-5 is slow: Use GPT-4-turbo
3. If ElevenLabs fails: Pre-recorded audio
4. If video gen fails: Static image + audio
5. Nuclear option: Pre-recorded demo video

📊 Metrics to Show

• Generation time: ~8 seconds
• Transformation options: 10+ formats
• Information preservation: 94%
• Potential market: 2M+ creators
• Revenue model: $50/month subscription

🔑 Environment Variables

# LLMs
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

# Audio/Video
ELEVENLABS_API_KEY=...
REPLICATE_API_KEY=...

# Scraping
BROWSERLESS_API_KEY=...
BRIGHT_DATA_PROXY=... # Backup for LinkedIn

# Demo Config
DEMO_MODE=true
CACHE_ENABLED=true
TARGET_AUDIENCE=yc # or 'afore'
MAX_GENERATION_TIME=10 # seconds

# Optional
REDIS_URL=redis://localhost:6379
SENTRY_DSN=... # Error tracking

🏃‍♂️ 12-Hour Timeline

Hour	Focus	Deliverable
0-2	Setup & Scraping	Working URL → text extraction
2-4	LLM Integration	Profile analysis + lyrics generation
4-6	Audio Pipeline	TTS + beat mixing working
6-8	Basic Frontend	Input → Results display
8-9	Video Generation	Basic video output
9-10	Transformations	Show any2any capability
10-11	Polish & Debug	Fix critical issues
11-12	Demo Practice	Rehearse, prepare backups

🎯 Demo Flow (90 seconds)

1. Hook (0-15s)

   • "We transform any content into any format"
   • Input judge's LinkedIn URL

2. Magic (15-45s)

   • Show real-time generation
   • Play diss track (keep it short!)
   • Audience laughs

3. Business (45-75s)

   • "But watch this..."
   • Transform to different formats
   • "Same engine powers content for 2M creators"

4. Close (75-90s)

   • "$50/month, 40% margins"
   • "Already have 100 beta users"
   • "This demo IS our marketing"

🐛 Known Issues & Solutions

Issue	Solution
LinkedIn requires login	Use cached profiles for demo
GPT-5 too slow	Fallback to GPT-4-turbo
Rate limits	Multiple API keys, rotate
CORS errors	Proxy through backend

🚀 Post-Hackathon

If we win:

1. Polish the transformation engine
2. Add more input sources (TikTok, Instagram)
3. Add more output formats
4. Build creator dashboard
5. Launch on Product Hunt
6. Sell to Adobe for $100M 😎

📝 Notes

• Keep it simple: Better to have 3 working features than 10 broken ones
• Demo is everything: 50% coding, 50% preparing demo
• Have backups: Everything that can fail, will fail
• Energy matters: High energy presentation wins

---

Built with 💔 and 🔥 at YC Agents Hackathon "Transforming content and feelings since 2024"