What it does: Grounds an LLM’s responses in your curated library of text documents.
Media: Text

Retrieval-augmented generation (RAG) is a framework that connects an existing large language model (LLM) to a knowledge base you curate. Instead of retraining or fine-tuning, you hand it your own resources and the LLM grounds every response in your curated library.

Workings

1. Collect documents

• Choose what you would like to ground your LLM in: research notes, dream diaries, film scripts, poems, archives etc.

2. Build a knowledge base

• Embed them into a searchable index so the system can retrieve relevant passages.

3. Connect to an LLM

• Run it locally (e.g., with Ollama).

4. Generate responses

• The LLM draws only from your curated library, grounding its answers in your sources.

5. Extend creatively

• Use the responses as inputs for other text-based generative processes or interactive events.

Why work with RAG?

• Shape the LLM by curating its knowledge base without any training process.
• You get grounded answers tied directly to your selected resources.
• It's private and offline when connected with a local LLM.

You get:

• Your own continuously updatable, locally running mini-LLM, shaped by the collection of documents you curate.