How to Make Smart Architecture Decisions when Building Gen AI Apps
summary
A practical recipe for building GenAI apps that uses Amazon Bedrock, serverless glue (S3/CloudFront, Cognito, API Gateway, Lambda, Step Functions) and a vector database for retrieval augmented generation to avoid fine‑tuning. Concrete takeaways: chunking + embeddings for semantic search, measure context recall/relevance, track faithfulness and answer relevance, and throttle/model‑cost tradeoffs (orders of magnitude cost differences). Demo use cases: customer Q&A via RAG works and gives sources; text‑to‑SQL agents are risky, slow and require strict validation and least‑privilege controls.
AI-written summary · about
chapters (9)
- 0:00Intro
- 1:36Generative AI
- 11:40Use case 1
- 19:56Objective 1: Fast
- 22:08Objective 2: Accurate
- 35:22Objective 3: Safe
- 37:52Use case 2
- 45:41Advice Summary
- 48:02Outro
related talks
29mCustomizing genAI applications for your business
19mMind your business: Secure your generative AI application on AWS
20mBuilding a secure end-to-end generative AI application in the cloud
07mBuild Generative AI Apps for Enterprises with Cloud
38mGenerative AI - Architectures and applications in depth
01h 02mArchitecture Patterns for AI-Powered Applications
28mImproving responses in generative AI applications
45mServerless & Event-driven Patterns for GenAI