Talk: A live, end-to-end demo of wiring an open-source vision-language model (SmolVLM) into Vertex AI - with a lightning primer on what Vertex AI is, which quotas matter (GPUs), and how to pick the right model tier for your latency x cost sweet spot. We’ll then drive that endpoint from a Firebase web app that streams camera frames and spits back analytics in milliseconds - real-time video AI minus the heavyweight MLOps baggage. Target Audience: Cloud and DevOps engineers, full-stack developers, AI-ML hobbyists, and startup builders already shipping (or keen to ship) on GCP who want a pragmatic path to weaving generative/computer-vision AI into their products and pipelines. Takeaways: A checklist for matching model size/tier to performance and budget. A lightweight pattern for streaming video to that endpoint via Firebase and turning frames into instant insights.