From RAGs to riches: Build an AI document interrogation app in 30 mins
summary
A working RAG pipeline gets built end to end: Ragna OSS framework wires embeddings, vector stores and LLMs; a Panel web app provides the chat UI; models include OpenAI GPT variants, Anthropic, a local GPU LLM (aeroborus) and vector DBs like chroma vs LanceDB. Live demo is flaky but a pre-run works, revealing chunking, embedding and similarity choices, a chroma bug, GPU load spikes and occasional hallucinations — practical failure modes and concrete knobs.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:07Speaker and demo setup
- 3:58Live demo with Tesla GM Ford PDFs
- 5:43AI adoption struggles
- 6:49What are LLMs
- 9:16How RAG works
- 13:53Introducing RAGNA framework
- 17:06Setup models and embeddings
- 18:55Chat widget with Panel
- 24:13Demo to production tools
- 27:20Open source RAGNA future
- 28:46Q&A chunking costs
- 30:08Handling inadequate search and vector databases
- 31:10LLMs for SQL data
- 32:39Fine-tuning vs RAG
- 33:47Hardware for fine-tuning
- 34:08Tables and figures in docs
- 35:35Summarizing documents
- 37:28Fine-tuning vs RAG clients
- 38:54Proprietary model risks
- 40:29Privacy cloud vs local
related talks
27mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
02h 40mFrom RAGs to riches: Build an AI document inquiry web-app
01h 13mBuilding Production-Ready AI Systems in 90 Minutes
01h 16mForget RAG Pipelines—Build Production Ready Agents in 15 Mins
42mTake Your RAG To The Max!
18mLLMs and GenAI: A Guide to Building Your Own RAG Model for Document Interaction
48mGoing Beyond Prompt Engineering
40mRolling out Enterprise AI: Tools, Insights, & Team Empowerment