Harnessing EDA & Workflows to Build Real World GenAI Apps
summary
An event‑driven review response service is built using Amazon EventBridge and AWS Step Functions to orchestrate toxicity checks with Amazon Comprehend, sentiment classification and response generation via Amazon Bedrock models (anthropic CLA 2 for sentiment, anthropic Claud 3 Haiku for generation). A 30k token monolithic prompt that took 45 seconds is decomposed into smaller calls that run in parallel, totaling 7 seconds and about 85% cheaper. Human reviewers use a wait for callback task token to approve or reject flagged outputs; the demo detects toxicity and resumes correctly.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:59Agenda
- 2:21Generative AI
- 11:39Event-driven Architecture
- 22:27Prompt chaining: Building out the workflow
- 33:56Human in the workflow
- 36:53Demo
- 42:02Resources
- 42:27Outro
related talks
36mHow we built an AI-powered code reviewer in 80 hours
31mLamRAG: From Data to Constructive Insights using Amazon Bedrock
38mBuilding AI responsibly with a GRC strategy
01h 02mArchitecture Patterns for AI-Powered Applications
45mPanel Discussion on AI for Code Generation
54mAI is Coming - Now What? Architecting the Future of Software
28mImproving responses in generative AI applications
27mBuilding Generative AI features