Vectorizing Images with LLMs [Full Stack Deep Dive]
In this video, we'll walk through how to build a complete end-to-end image vectorization and similarity search system using modern AI tools. We'll explore how to generate image embeddings with a multimodal LLM, store them efficiently in ChromaDB, and see how LLM microservices power the semantic search logic—all while checking out the architecture and watching a live demo in the browser.
Description
In this video, we walk through how to build a complete end-to-end image vectorization and similarity search system using modern AI tools, multimodal LLMs, and GPU-powered infrastructure.
🔍 What You’ll Learn
• How anything can be represented as a vector — including images
• How to generate image embeddings using an OpenAI multimodal model
• How to store vectors efficiently using ChromaDB, a high-performance vector database
• How LLM microservices (running on vLLM) power the semantic search logic
•A deep dive into the architecture, design decisions, and multi-service setup
• A full demo of the final solution live in the browser
You’ll see exactly how these pieces fit together to deliver an end-to-end multimodal AI experience.
💻 Source Code:
Source code from the video is available on GitHub for review, reuse, and extension:
https://github.com/robkerr/robkerrai-demo-code/tree/main/dgx-spark-image-vectorization
0:00 Introduction
033 Demo
1:38 Architecture
3:40 Model Card
4:00 Kaggle Dataset
5:01 What are Vectors?
6:02 Vector DB Loading
7:15 Embedding Model Code
10:00 Run the Loading Script
11:08 Re-test the web app
11:42 Summary