Building Production-Ready AI Systems in 90 Minutes
summary
Builds a working RAG pipeline in 90 minutes using LangChain plus ChromaDB and OpenAI embeddings from the repo; loads a PDF, splits into chunks, generates embeddings, and wires a retriever → generator chain that returns answers with sources. Demonstrates gaps: keyword vs semantic search, chunking choices that lose context, embedding dimensional tradeoffs, and an embarrassing env-var mismatch that broke production tests. Adds evaluation tests (dEval) and guardrail patterns, shows what was implemented, what failed, and which parts become separate services in real deployments.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 15:33Data loading pitfalls
- 19:33Document splitting strategies
- 26:23Embedding models nuances
- 37:21Vector database setup
- 41:44Semantic vs keyword search
- 49:40Prompt engineering and guardrails
- 59:36Testing AI performance
- 1:05:42Deployment lessons & Q&A
related talks
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
01h 16mForget RAG Pipelines—Build Production Ready Agents in 15 Mins
28mImproving responses in generative AI applications
48mBuilding AI Powered Applications with Java
48mGoing Beyond Prompt Engineering
27mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
21mHow RAG is Redefining AI Capabilities
01h 28mFrom RAG to Riches Building Enterprise Search Systems with LLMs