👋 Introduction Hey builders, creators, and AI tinkerers, I’m Maxime Yao, an 20-year-old Business student at the University of Exeter and a Product Manager intern at Amazon. For the past two months, I’ve been building AceClip.com solo AI-powered system that turns long-form videos (podcasts, interviews, lectures) into short, searchable, personalised knowledge clips. I’m now looking for a technical co-founder to help take AceClip from a working MVP to a production-ready web platform. If you’re passionate about AI, education, and the future of knowledge discovery, this might resonate with you. 🎬 What Is AceClip? AceClip is your “personal YouTube brain.” It helps users find, organize, and revisit the smartest moments buried inside hours of video content. Core functions: - 🧠 Automatically detects the most insightful moments in podcasts & interviews - ✂️ Generates professional short-form clips with captions, face cropping, and speaker tracking - 🔍 Supports semantic video search using RAG + vector embeddings - 📚 Builds a personalized knowledge library an encyclopedia of ideas tailored to you Think of it as the next evolution of intentional learning: turning internet noise into structured, actionable wisdom. ⚙️ How It Works – Under the Hood - Language & video stack: Python | OpenCV | FFmpeg | GPT for semantic understanding - Perception & audio: InsightFace for facial recognition, pyannote.audio for diarization - Knowledge layer: RAG + embeddings for search; DeepSeek-OCR for 10× transcript compression - Performance: Processes 1 hour of video in ≈ 15 minutes on cloud GPUs - Scalability: Embedding billions of minutes for <$100 one-time processing and $10–20 monthly storage - Codebase: ~15 K modular lines (AI-assisted); local backend fully automated at MVP stage The system chunks transcripts into 3–7 minute segments (≈1 000–1 500 words), indexes metadata (speakers, timestamps, topics), and enables natural-language queries across millions of clips.