Loading video player...
Gemini Embedding 2 just dropped, and this release is a big step for multimodal retrieval workflows. In this video, I walk through what is new, key limits to know before you build, benchmark highlights, and a practical ChromaDB demo where we embed mixed media and run semantic queries against one shared vector space. --- š§āš» Notebook: https://colab.research.google.com/gist/alejandro-ao/d74d07930e213188e9e9db7da7e9748d/gemini-embeddings-2.ipynb --- š¤ *Topics Covered* - What Gemini Embedding 2 adds vs prior embedding workflows - Input limits and practical constraints (text/image/video/audio/PDF) - Benchmark highlights and how to interpret them - Building a multimodal vector database with ChromaDB - Querying and validating retrieval quality in a real notebook ## Connect with me - X: https://x.com/_alejandroao - LinkedIn: https://www.linkedin.com/in/alejandro-ao/ --- ā±ļø *Timestamps* 0:00 Intro + what Gemini Embedding 2 is 0:32 Capabilities, constraints, and benchmark context 3:54 ChromaDB implementation + retrieval demo