Building Performant RAG Applications for Production
summary
Builds a support-ticket RAG pipeline that moves from a single-product baseline to production patterns: uses llama index plus a Qdrant-like vector store (transcript spells it "quadrant") with hybrid search (dense + sparse embeddings) for retrieval, generates per-chunk questions via an LLM and stores them as index entries (summary indexing), adds metadata filters and LLM-driven query expansion, and implements query routing and API-backed dynamic context (user plan lookup) so the system selects the right pipeline(s) and fuses results into a final draft.
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:50Agenda
- 1:42Why use RAG?
- 3:52Performant RAG?
- 4:44Use-case
- 5:18First iteration: The simple case
- 7:27Demo
- 9:48Second iteration: Multiple categories of documentation
- 12:50Demo
- 14:43Third iteration: Unstructured documentation
- 17:50Demo
- 19:56Fourth iteration: Dynamic context & actions
- 24:22Demo
- 29:15Take-aways
- 31:33Outro
related talks
39mWhat if you could build production-ready RAG entirely in Elixir?
30mI Want to Build a RAG System, Now What?
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
11mBuilding easily deployable RAG applications with Llamastack
02h 47mFrom naive to advanced RAG: the complete guide
33mRAG Pipelines Letting You Down? Fitch Group's Solution for Financial Documents
37mBuilding advanced RAGs with Elasticsearch OpenSearch
01h 12mBuild RAG using Llama 3.1