Generative AI Text and Multimodal Embedding Models for Real World Use Cases
summary
Shows how Vertex AI text and multimodal embeddings power real systems with concrete numbers and tools: 768‑dim text vectors, ~1,148‑dim multimodal vectors, stored in Google Cloud Vector Search and indexed with ANN (ScaNN) for fast nearest‑neighbor. Demos index 8,000 Stack Overflow Q&A and run image search over 5.7M images returning ~60 similar items in milliseconds. Uses Two Sigma + Google Cloud evaluation on the Massive Text Embedding Benchmark (MTEB); fine‑tuning yields ~8 percentage points improvement (up to ~40% relative gain).
AI-written summary · about
chapters (7)
- 0:00Introduction to embeddings
- 2:28What are embeddings
- 8:25How embeddings work under the hood
- 14:55Live demos of embeddings
- 22:48Embeddings in financial services
- 27:19Choosing the right embedding model
- 31:10Fine-tuning embeddings
related talks
33mEnter the Brave New World of GenAI with Vector Search
48mEnter the Brave New World of GenAI with Vector Search
34mEnter the Brave New World of GenAI with Vector Search
31mGoogle's AI Map
09mBuild generative apps faster with Vertex AI
31mExploring LLMs across clouds – A Year in the Life
57mUsing Vector Databases for Multimodal Embeddings and Search
03mContextual understanding using Generative Artificial Intelligence