Descriptions:
Nate Herk walks through two working applications built with Google’s newly released Gemini Embedding 2 — the company’s first natively multimodal embedding model — combined with a Pinecone vector database and Claude Code as the implementation assistant. The video is both a product introduction and a practical build tutorial.
The first demo ingests a 68-page vacuum cleaner PDF manual into a multimodal Pinecone database, storing both text chunks and embedded images. A chat interface built on top retrieves relevant results with page citations, confidence scores, and actual diagrams from the source document — something traditional text-only RAG pipelines struggle to deliver cleanly. The second demo uses 13 roofing project images to build a visual similarity search tool: uploading a roof photo returns matched past projects with metadata including cost ranges, team size, and job duration.
Herk then provides a step-by-step setup guide using Claude Code inside Visual Studio Code. He shows how plan mode, combined with Google’s embeddings API documentation URL pasted directly into the prompt, allowed Claude Code to scaffold the full project — `.env` file structure, Pinecone database configuration, and application logic — with minimal manual coding. Required credentials are a Pinecone API key, Gemini API key, and OpenRouter API key. The tutorial is accessible to developers with no prior RAG experience and is particularly relevant for anyone building document or image retrieval systems.
📺 Source: Nate Herk | AI Automation · Published March 11, 2026
🏷️ Format: Hands On Build
