LLM-Audited EHR Pipelines: 87% Accuracy with 70% Fewer Resources
summary
Claims an LLM‑based auditing pipeline that reaches 87% detection rate while using 70% fewer resources, moving pipeline coverage from 35% to 92% and cutting manual hours from 2,400 to 720 annually. Implements a multi‑layer architecture with readonly API extracts, an LLM orchestration service, hierarchical prompting and chain‑of‑thought workflows that find clinical logic bugs (sepsis calculations, med reconciliation, outdated protocols) and integrates with CI/CD pipelines for near‑real‑time checks via hierarchical prompting.
AI-written summary · about
chapters (12)
- 0:00Introduction and Research Overview
- 0:31The Problem with Current Healthcare Data Pipelines
- 1:52Limitations of Traditional Auditing Approaches
- 3:07Introducing the LLM Powered Framework
- 4:01Framework Architecture and Components
- 5:16Advanced Prompting Strategies
- 6:23Implementation Results and Impact
- 7:32Case Studies and Real-World Applications
- 8:37Comparative Performance Analysis
- 9:29Ethical Considerations and Integration
- 10:27Future Directions and Broader Applications
- 11:23Conclusion and Final Thoughts
related talks
31mLLM Enabled Multi Document Correlation for Financial Compliance
38mBehind the Firewall: Designing Secure LLM Pipelines for Regulated Data
44mProduction-Grade LLMs & AI Agents
21mTrust, but Verify: Lessons from Deploying LLMs in a Large Health System
38mMaking LLMs reliable - A practical framework
24mLLMs in Regulated Industries: Challenges and Governance Solutions
50mHow to Test LLM Agents
29mCustomizing and Evaluating LLMs, an Ops Perspective