🧠 Build Your Own Oracle: Local AI + RAG on your Library

​What if your chatbot only answered from docs you trust?

​Generic cloud chatbots freestyle from public training data and send every prompt to a vendor. We talk about the other path: self-sovereign AI built around RAG (retrieval-augmented generation). You run a local model, keep a corpus you curate, and at question time the system finds the relevant passages and answers from those. Accuracy for your domain, sources you control, and privacy when you want the conversation to stay on your hardware.

​Night in three parts

​1. Basics The building blocks: a local model (we’ll talk options that fit real home hardware, not only one brand), Open WebUI as the chat front end, and system directives so the model has a job and rules. Live walkthrough of the simple path.

​2. Advanced (if you want it) RAG done well, evaluation harnesses, and techniques for steering the model so it stays on your material instead of inventing. High-level map of what’s possible once the basic stack works.

​3. Showcase Something I actually run: specialized oracles grounded in real corpora. Demo, not a sales pitch.

​Who it’s for

​Builders, home-lab people, anyone tired of “it doesn’t know my docs.” No PhD required. Beginners get the components; people already on local models get RAG, steering, and the showcase.

​Bring questions. Bring a domain you care about.