2126 - 利用 InstructLab 和 AI 模型调优革新紧急医疗服务.pdf

1、Orlando,FLOctober 69IBM TechXchange 2025Session code(2126)Daniel SteinDaniel SteinKPMG,Senior Data ScientistSyeda AmeenaSyeda AmeenaIBM,AI EngineerRevolutionizing Emergency Medical Services Using InstructLab-AI Model TuningAgenda1.Context Who is Hato Hone St John&why this matters2.The Challenge The

2、scale,gaps,and stakes of clinical auditing3.The Decision Why Sovereign AI,not off-the-shelf AI4.How We Built Sovereign AI From taxonomy to clinical validation using InstructLab5.Closing What this means for youIBM TechXchange|2025 IBM Corporation21.700,000+patient interactions every year at Hato Hone

3、 St John2.New Zealands frontline ambulance and emergency service3.Trusted with lives every time someone dials 111111This Story is About People,Not Just TechnologyThis Story is About People,Not Just TechnologyIBM TechXchange|2025 IBM Corporation3The Challenge The scale,gaps,and stakes of clinical aud

4、itingIBM TechXchange|2025 IBM Corporation41.430,000 electronic patient report forms(ePRFs)each year2.2.Contractual auditsContractual audits:520%up to 7,000 reviews monthly3.Resource:250 volunteers+1.5 full-time staff4.Reality:Only 5%records+1%conversations reviewed430,000 ePRFs per year 7,000 requir

5、ed audits monthly 5%actually auditedSt John-ePRFIBM TechXchange|2025 IBM Corporation5Clinical Practice Guidelines FallsIBM TechXchange|2025 IBM Corporation6Clinical auditing requires:Clinical auditing requires:1.Contextual understanding2.Consistency in applying rules3.Transparent reasoning clinician

6、s trustHow We Built Sovereign AI From taxonomy to clinical validation using InstructLabIBM TechXchange|2025 IBM Corporation7Start Small,Where It Mattered Most.Began with critical risk factor critical risk factor cases high-risk,real-world impactIBM Granite LLM+Clinical Review Manager at Hato Hone St