Omniversal AI

• 

  Chat page

• 

  Landing page / Home page

• 

  Features page

• 

  Pricing page

• 

  Login

• 

  Form modal after log in

Inspiration

We wanted to build something that combined cutting-edge AI APIs (Gemini 2.5 Flash + ElevenLabs) with a satirical commentary on the AI hype machine. The idea: make a product that looks 100% enterprise-grade but exists purely to roast the user.

What it does

Omniversal AI presents itself as a "Productivity Platform" but delivers sophisticated roasts in multiple personas (Shakespearean, Gen Z, Corporate CEO). It analyzes user input, finds logical/technical flaws, and roast them with context-aware humor—then speaks it aloud via ElevenLabs real-time voice synthesis.

How we built it

• Frontend: Next.js 16 + Tailwind with a slick dark-mode UI and fake analytics dashboards
• Backend: Gemini API with heavy system prompting for persona preservation
• Voice: ElevenLabs TTS with auto-play for instant roasts
• Deployment: Vercel (Edge-optimized for speed)

Challenges we ran into

• Vercel function timeouts on free tier with longer audio generation
• Persona drift (keeping the AI in-character across multi-turn conversations)
• Balancing "funny" with "not actually malicious"

Accomplishments that we're proud of

• Working multi-persona modular engine
• Real-time audio delivery with zero user interaction required
• Fully deployable MVP that actually roasts

What we learned

• System prompting is an art form for persona consistency
• Edge functions require careful timeout management
• The "fake analytics" gag adds significant comedic value

What's next for Omniversal AI

• Add more personas (Film Noir, Meme-Lord, Corporate HR)
• "Neural Behavioral Report" email dispatch feature
• Leaderboard for "Most Roasted" community member

Built With

• elevenlabs
• framermotion
• geminiapi
• lucidereact
• next.js
• nodemailer
• tailwindcss
• vercel