{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/748094b8ddd34428adf5c1b02a98ccf4\" frameborder=\"0\" width=\"1280\" height=\"960\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":960,"width":1280,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":960,"thumbnail_width":1280,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/748094b8ddd34428adf5c1b02a98ccf4-bdd766d7d2fd0e3f.gif","duration":276.702,"title":"Explaining Context Isolation with Brand Summaries","description":"This Loom explains how the system isolates conversational context per brand using a stored brand summary instead of full chat history. The speaker describes updating a target audience and other keywords such as young women and young men, and then using the current brand summary fetched from the database to generate a professional response and limit the summary on each request. They note two main design decisions: using PostgreSQL instead of in-memory storage because data is lost on server restart, and prompting the model to return JSON. The flow also updates the brand summary back to the database for future requests, keeping conversations for different brands isolated."}