the-ai-merge/multimodal-agents-course ? reverse-engineered prompt

Build me the Kubrick multimodal AI agent project from this repo, with a clean local setup I can run end to end. I want the system to be able to take video, images, text, and audio, then let an agent search and reason over that content through MCP. Include the three parts I can see here, the backend API, the MCP server, and the UI, and make sure they work together smoothly.

Use the course style and HAL themed branding from the README, but keep the app practical and polished. I’d like sensible defaults for local development, clear startup instructions, and a working demo flow that shows the agent answering questions about uploaded or indexed media. If anything depends on current docs or APIs, look them up online and wire it up the right way. Also make sure the project is easy to run with Docker or a simple local dev command, and that the multimodal pipeline and observability pieces are included where they belong.