Serverless & Event-driven Patterns for GenAI
summary
Shows concrete serverless patterns on AWS for GenAI: use Amazon Bedrock or SageMaker JumpStart for models, fronted by API Gateway + Lambda for prompt enhancement and model abstraction, use SQS buffering to protect low‑scale model endpoints and deliver async results via WebSockets or polling, implement retrieval augmented generation with vector DBs (OpenSearch Chroma Pinecone) for domain knowledge, and run scheduled Step Functions that create SageMaker endpoints, process 5,000–100,000 documents with distributed map, then delete endpoints to save cost.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:52Agenda
- 1:40Introduction to generative Al
- 7:26Use cases
- 11:36Why use serverless with generative Al?
- 15:01Serverless services
- 18:52Serverless patterns for generative Al
- 30:11How to build context based on internal knowledge?
- 37:08Hosting foundation models
- 37:46Documents summarization on SageMaker at scale
- 43:12Summary
- 44:09Outro
related talks
15mServerless & Agentic AI: Better Together
48mHow to Make Smart Architecture Decisions when Building Gen AI Apps
32mBuilding a Serverless GenAI Platform
23mBuilding Agentic AI Systems with AWS Serverless
47mOutsmarting the Genie: Smart Architecture Decisions for GenAI Apps
01h 00mSecurely accelerating generative AI innovation
07mServerless: Accelerate your AI
30mStreamlining data governance