Orchestrating Agentic State Machines with LangGraph | Rajeshwari Sah | Conf42 ML 2026 скачать в хорошем качестве

Orchestrating Agentic State Machines with LangGraph | Rajeshwari Sah | Conf42 ML 2026

Read the abstract ➤ https://www.conf42.com/Machine_Learni... Other sessions at this event ➤ https://www.conf42.com/ml2026 Apply to speak ➤ https://cfp.ninja/?q=conf42&status=open Join Discord ➤   / discord   Chapters 00:00 Intro: Orchestrating Agent State Machines for Clinical Documentation 00:15 What “Agentic” Means (Engineering View): Nodes, State, and Routing 00:50 The Clinical Scribe Use Case + 4-Agent Pipeline Overview (SOAP) 02:41 Why This Matters: Documentation Burden, Burnout, and System Impact 04:13 Why Not One Big Prompt: Accuracy, Auditability, and Debuggable Workflows 05:36 Why Clinical Notes Are Hard: Messy Dialog, Negation, Time, Uncertainty 07:56 What We’re Building: Inputs/Outputs, Pipeline, and Two Key Design Choices 11:01 Why LangGraph: State Machines for Branching, Retries, and Traceability 13:38 LangGraph Vocabulary Crash Course: State, Nodes, Edges, Routers, Cycles 15:34 The Shared MedicalScribeState: Audit Logs, Errors, and Testability 20:39 Schemas & Structured Extraction: Pydantic Models as an API Contract 22:27 Node Implementation Template + Transcription Agent Responsibilities 24:58 Extraction Agent Deep Dive: Validation Retries and Negation Handling 29:34 Negation, temporal qualifiers & deterministic normalization in extraction 31:04 From structured encounter to SOAP note: summarization constraints & editable sections 32:15 Reducing hallucinations: structured inputs, no-new-facts prompts & low temperature 34:33 Consistency agent: catching omissions, contradictions & invented facts 35:27 Consistency checks taxonomy: hard rules vs semantic judging vs safety scans 37:56 Graph orchestration with LangGraph: explicit routing, safe failures & debugging 41:09 End-to-end demo walkthrough: transcription → extraction → SOAP → consistency 47:54 Retrieval grounding: mapping diagnoses to ICD-10/SNOMED as a pipeline node 48:32 Measuring quality per agent: datasets, metrics & iteration strategy 51:07 Production readiness: PHI, redaction, prompt injection, monitoring & human review 53:54 Setup, key takeaways & next steps (feedback loop, better grounding)