Outsmarting the Genie: Smart Architecture Decisions for GenAI Apps
summary
Practical guide to architecting GenAI chatbots on AWS with concrete patterns: static site on S3/CloudFront + Cognito, API Gateway → Lambda, and model access via AWS Bedrock; use a vector database and retrieval augmented generation (RAG) pattern with embeddings for FAQ accuracy. Measures include context recall/relevance, faithfulness and answer relevance; guardrails use prompt engineering, input/output validation and a secondary LLM for safety. Tradeoffs called out: model size versus latency and cost, monitoring for model-DoS and chunking/indexing tweaks; examples show RAG plus prompts works and needs orchestration and testing.
AI-written summary · about
chapters (9)
- 0:00Intro
- 4:58Generative AI
- 14:01Use case 1
- 15:09Objective 1: Accurate
- 32:39Objective 2: Safe
- 35:12Objective 3: Fast
- 36:35Use case 2
- 44:50Advice Summary
- 46:52Outro
related talks
33mCompose GenAI Apps: May the Force Be with Your Data and Models
36mGENAI-Powered chatbot to increase team efficiency
19mMind your business: Secure your generative AI application on AWS
38mCraft your own Generative AI Chatbot with Amazon Bedrock
29mCustomizing genAI applications for your business
45mServerless & Event-driven Patterns for GenAI
01h 02mArchitecture Patterns for AI-Powered Applications
44mBuilding a Domain Specific GenAI Chatbot with Serverless