Improving responses in generative AI applications
summary
Demonstrates a concrete RAG pipeline using Amazon Bedrock to auto-discover data sources, ingest documents, run embedding models into a vector store (default OpenSearch or bring your own like Pinecone), and call a retrieve and generate API. Shows an enterprise bid‑writing demo where default fixed chunking fails to fetch a specific safety statistic but semantic chunking retrieves the exact citation, proving the method works. Covers multi‑query + reranking, prompt augmentation, token optimization, guardrails, and four evaluation dimensions: retrieval, generation, UX, and performance.
AI-written summary · about
chapters (13)
- 0:00Introduction and welcome
- 1:33Limitations of large language models
- 3:17Approaches to improve responses
- 4:21What is RAG?
- 6:27RAG architecture patterns
- 9:36Best practices for high-quality RAG
- 11:22Vector database hydration
- 14:11Retrieval strategies
- 16:38Prompt augmentation
- 19:24Response generation and safety
- 21:53Evaluating RAG performance
- 23:15Demo use case: Bid writing
- 27:25Recap and closing
related talks
38mGenerative AI - Architectures and applications in depth
44mEnhance Your Generative AI Applications with RAG
50mRetrieval Augmented Generation Beyond Toy Demos
21mHow RAG is Redefining AI Capabilities
37mBuilding Generative AI Applications
19mMind your business: Secure your generative AI application on AWS
25mAdvanced RAG Systems: Scaling Retrieval-Augmented Generation in Production
20mBuilding a secure end-to-end generative AI application in the cloud