Video Highlight and Best-Moment Auto Extraction

You are a sports and live-event video AI engineer. Build a system that automatically detects and extracts highlights from long-form live and recorded video for quick-turnaround clips. Inputs: - Source content: [INSERT - sports, esports, concerts, conferences, hours] - Highlight types: [INSERT - goals, knockouts, punchlines, applause, Q&A] - Output channels: [INSERT - vertical social, app push, OTT recaps] - Turnaround target: [INSERT - seconds/minutes after the moment] - Available signals: [INSERT - score feed, chat, audio, closed captions] Deliver in modules: 1. Objective: ranked highlight segments delivered as auto-reframed, captioned, brand-ready clips. 2. Process: multimodal detection (action recognition, audio energy, crowd reaction, caption cues, external data feeds) -> boundary detection and ranking -> auto-edit with replays/slo-mo -> captions and graphics -> publish. 3. Toolchain: [INSERT - action model, audio analysis, clip editor, graphics engine], live capture, publishing API. 4. QA: false-positive review, rights and blackout windows, brand-safety filters, clip quality and caption verification. 5. Delivery: clip library with tags/rank/timestamps, per-platform aspect variants, and view/CTR analytics. Output: detection-model plan, latency budget, highlight JSON schema, and cost per event. ================