I want to build a YouTube Live Stream Content Assistant for my AI/automation channel. Goal: Before I start a live stream, the system should scan selected English AI/automation YouTube channels, detect unusually high-performing videos, and prepare a Turkish live stream briefing for me. Context: I create Turkish live streams about AI, automation, n8n, Claude Code, OpenAI, Anthropic, AI agents, self-hosting, coding assistants, and practical AI workflows. I do not want a generic news tracker. I want a system that finds videos that are performing unusually well compared to each creator’s own normal performance. The goal is to discover what topics are currently getting abnormal attention in the English AI YouTube ecosystem and turn them into live stream ideas. MVP Requirements: 1. Channel tracking Create a configurable list of YouTube channels focused on AI, automation, AI agents, AI coding, n8n, Claude Code, OpenAI, Anthropic, and related topics. Example channels: - Matt Wolfe - The AI Advantage - All About AI - Matthew Berman - Wes Roth - AI Explained - Nate Herk / automation-focused creators - n8n-related creators - Claude Code / coding agent creators - OpenAI / Anthropic / AI tools commentators The channel list should be editable from: config/channels.json 2. Video collection For each channel, fetch the latest 10 to 20 videos. For each video collect: - Channel name - Channel ID - Video title - Video URL - Publish date - Video age in hours - Video age in days - View count - Views per hour, VPH - Thumbnail URL - Duration, if available - Subscriber count, if available - Like count, if available - Comment count, if available Do not fetch transcripts in Phase 1. Do not summarize full videos in Phase 1. Focus first on titles, view counts, publish dates, velocity, and outlier detection. 3. Outlier detection model Build an outlier detection model inspired by the public vidIQ-style logic. The model should not only check raw views. It should compare each video against the same channel’s normal performance.