{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/3fd8b84a3e8d48e8ba8d7e29f6deef21\" frameborder=\"0\" width=\"1874\" height=\"1405\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1405,"width":1874,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1405,"thumbnail_width":1874,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/3fd8b84a3e8d48e8ba8d7e29f6deef21-8cdff6568c52e074.gif","duration":341.365,"title":"Hindsight: AI Insights for Instagram Creators","description":"This Loom demonstrates Hindsight, a tool that gives Instagram creators specific content advice based on their own posts. It uses LLM agents to scrape a creator’s Instagram content, then analyzes outcomes using available metrics such as views and shows which features hurt or help, along with inconclusive signals. In the demo, for Varun Maya (1.3 million viewers, 89 recent posts), educational tone was linked to lower views, while bold claims and evening posting showed better performance. It also compares creators like Varun Maya and NASA, noting educational content helped NASA’s engagement where it hurt Varun Maya, and it provides five next-post suggestions and a personal run showing carousels hurt while Reels performed best. The speaker explains the architecture using Amplify for scraping, LLMs for feature normalization, and a FastAPI plus Next.js UI."}