AI

"AI" in scare quotes because it's not really AI yet.

Basics

• https://www.fast.ai/ - Learn AI

LLMs

• Ollama for running models locally

RAG

• https://vectorize.io/multimodal-rag-patterns/
  • Basically, store vector encodings in the same vector space across all content types.
  • Pattern 3 in the above is about storing pointers to the raw file for later retrieval when your LLM is multimodal and needs access to the source data
  • If you don't need the source data, you can use Pattern 1, which is to just encode a text description. e.g. You take the image and run it through a "describe this image" and then use the resulting text for your encoding.

Agentic RAG

Other pages

• Pose Estimation
• Variational Autoencoder