Michael Limberger

Need me? Email mike@limberger.ca

2026-05-14 · 13 parts

This was a Techalicious talk about retrieval. We hand a local model a language it has never seen, and watch it answer from documents we gave it. No fine-tuning run. No cloud API. The pictures stay pictures; the write-up sits under each one.

Part 1

Teaching AI Something It Was Never Trained On

Part 2

Why use local AI?

Part 3

Why not fine-tuning?

Part 4

The research backed up the failures

Part 5

Retrieve, Augment, Generate

Part 6

Normal RAG

Part 7

OpenWebUI, the 10-minute RAG setup

Part 8

Where normal RAG quietly breaks

Part 9

With RAG Proxy

Part 10

Wiring the proxy in, one config change

Part 11

Technology stack

Part 12

More features (the parts we did not demo)

Part 13

Same interface. Everything smart happens upstream.