Agentic Ambient Active Listening
summary
Abridge runs an ambient clinical listener that ingests 500 ms audio chunks into Kafka, uses Flink to window and transcribe, and drives decision logic in Temporal with signals and child workflows to decide which products to run; system targets sub-5s turnaround for client responses, handles billions of actions/month across hundreds of workflows, and uses Nexus to avoid HTTP-retry storms. A proof-of-concept to‑do agent detects “take blood pressure,” waits for the spoken value, and marks the task complete.
AI-written summary · about
chapters (14)
- 0:00Introduction and speaker intro
- 0:53Speaker background and Abridge overview
- 2:07Problems with synchronous microservices
- 3:58Why Temporal solves these issues
- 4:40Path to workflow orchestrators
- 5:49Abridge’s encounter workflows
- 6:59Refactoring workflow anti-patterns
- 8:45Agentic Ambient Active Listening
- 9:30Streaming audio system design
- 10:37Temporal’s streaming limitations
- 11:40Agent workflow harness in Temporal
- 12:48Demo: to-do list agent
- 13:53Streaming tokens with Temporal
- 14:36Closing remarks and thanks
related talks
42mAgentic AI with AWS Streaming
31mBuilding Agentic AI Platforms
01h 01mAccelerate data insights with agentic data exploration
23mContact Center Voice AI: Low-Latency Intelligence Extraction from Messy Audio Streams
40mMulti-agent AI Systems
26mManufacturing Alerts To ReActive Agents: Self-Evolving AI Agents on K8S for Predictive…
29mAgentic AI: What It Is and Why It Matters
27mUsing GenAI with Audio Data for Driving Better Customer Experience