
TeamDiscovery is a multi-style AI video captioning platform that solves a universal content creation problem: rewriting the same video message for different platforms and audiences. Upload a short video (30–120 seconds) and our pipeline automatically generates four distinct, ready-to-publish captions — each with a unique tone: Formal (professional), Sarcastic (witty), Humorous Tech (developer-culture humor), and Humorous Non-Tech (relatable everyday comedy). Under the hood, the pipeline combines three AI models: Groq Whisper for audio transcription, Fireworks Vision for frame/scene analysis, and DeepSeek-V4-Pro for multi-style caption generation. Every output passes through a 14-rule validation system that catches AI reasoning dumps, instruction echoes, truncated words, and lyric leaks — with a 3-attempt auto-retry loop using progressive backoff. The frontend is built with React 18, TypeScript 5.9, Vite 7, and Tailwind CSS 4, featuring a custom cyberpunk design system with Framer Motion animations, particle-canvas backgrounds, 3D tilt cards, and neon glitch effects. The backend runs on Supabase with Edge Functions as secure API proxies — no AI keys ever touch the browser. Key features include per-style caption regeneration (fix one without re-running the full pipeline), console-based pipeline timing for performance transparency, and one-command Docker deployment. The project is fully open-source under the MIT license. Built for content creators, marketers, and social media teams who need to repurpose one video across LinkedIn, Twitter, TikTok, and Discord — without spending hours rewriting.
13 Jul 2026