NVIDIA-AI-Blueprints/video-search-and-summarization — reverse-engineered prompt

Build me a working video intelligence app based on this NVIDIA Video Search and Summarization blueprint.

I want to be able to upload or connect video footage, then ask plain English questions like “what happened near the loading dock” or “show me clips where someone entered after hours”. The app should return useful answers, summaries, reports, and matching clips when possible. It should also support longer videos by breaking them down and summarizing the important parts, plus basic real time alert style workflows where video streams can be checked for unusual activity or verified alerts.

Use the existing Python agent pieces, deployment setup, and the provided UI as the starting point. Keep it focused on the documented NVIDIA VSS workflows, video Q&A, search, long video summarization, clip retrieval, and alert verification. Make it easy to configure NVIDIA API keys and model services, with clear setup steps for local development. Look up the current NVIDIA VSS docs online if you need to.