What if you could build production-ready RAG entirely in Elixir? George Guimarães, ElixirConfEU 2026

Репост: https://www.youtube.com/watch?v=tNBQQSeuAkE Плейлист "ElixirConf EU 2026": https://www.youtube.com/playlist?list=PLvL2NEhYV4ZsIgecSexrk26--PBWiEyUy Arcana: What if you could build production-ready RAG entirely in Elixir? This talk was recorded at ElixirConf EU 2026. Check our upcoming events at https://elixirconf.eu/ -- -- -- What if you could build a complete Retrieval-Augmented Generation (RAG) system without ever leaving the BEAM? In this talk, George Guimarães introduces Arcana, an open-source library designed to make this a reality. By using Nx and Bumblebee, Arcana allows you to comfortably run embedding models locally, removing the need to rely on external APIs. Combined with pgvector for storage and Phoenix LiveView for a clear, responsive dashboard, you get a powerful RAG solution that feels completely native to Elixir. We will walk through the fundamentals of how RAG works and explore efficient document chunking strategies. You will also learn about building agentic pipelines for tasks like query expansion, re-ranking, and creating self-correcting answers. Join us to discover how pluggable components and a solid evaluation framework can help you measure and improve retrieval quality. You will leave with the practical knowledge needed to add semantic search to your Phoenix applications, keeping your AI stack simple, fast, and entirely built in Elixir. -- -- --