Agentic AI with AWS Streaming
summary
Shows a concrete pipeline: Kinesis or MSK ingest IoT telemetry, Apache Flink does stateful real-time feature generation and anomaly detection, anomalies post to SQS and trigger an agent-core runtime that calls Bedrock models and returns structured JSON decisions. Claims scalability with Kinesis 807 million records/sec peak and SQS 166 million events/sec peak, demos a geofence workflow that reduces false positives via playbook rules, and recommends batching, caching, cross-region failover, and decoupling agents with queues.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:14Streaming data foundations
- 6:12AI agents and streaming
- 12:09Anomaly detection patterns
- 17:18Building with AgentCore
- 20:28Multi-agent workflows
- 25:19Event-driven architecture
- 26:58Design considerations
- 29:47Geo-fencing alert demo
- 40:49Conclusion and takeaways
related talks
16mEnterprise AI Innovation: GenAI Experimentation
44mBest practices for analytics and generative AI on AWS(ANT302)
50mThe tech behind modern AI/ML use cases in Gaming
01h 01mAccelerate data insights with agentic data exploration
16mEnterprise AI Innovation: Gen AI Experimentation
26mEngineering Excellence: Platforms and Architectures for Real-Time AI Analytics
44mFlink Jobs as Agents – Stream Processing for Agentic AI
24mBuilding GenAI applications with Amazon Bedrock