Hybrid RAG–LLM Framework for Supplier Risk Intelligence
summary
Presents an implemented RAG+LLM pipeline that retrieves ERP records plus external news, regulatory and ESG signals, reasons over that evidence, and emits traceable risk narratives instead of opaque scores. Evaluations across manufacturing and tech networks show higher detection precision, faster response and broader coverage versus rule‑based systems, and the system supports continuous monitoring across thousands of suppliers with an audit trail per conclusion and built‑in bias mitigation and human oversight.
AI-written summary · about
chapters (14)
- 0:00Why Supplier Risk Isn’t Predictable Anymore (and What This Talk Covers)
- 0:58Where Traditional Supplier Risk Models Break Down
- 3:103 Reasons Legacy Risk Intelligence Falls Short: Static, Limited, Late
- 5:13The Solution: An Explainable Hybrid RAG + LLM Risk Intelligence Pipeline
- 7:23Data Integration: Blending ERP Context with External Intelligence Signals
- 9:30Why Retrieval-Augmented Generation Improves Early Risk Detection
- 10:47Real-World Evaluation Results: Detection, Speed, and Coverage Gains
- 12:30How Multi-Signal Risk Emerges: Financial + Compliance + ESG Together
- 14:26Auditability by Design: Provenance from Source Docs to Decisions
- 15:55Governance Controls: Bias Mitigation, Interpretability, Human Oversight
- 17:25Performance & Scale: Continuous Monitoring Without Slowing Down
- 18:44What You Get in Practice: Precision, Faster Response, Compliance, Scale
- 20:17Wrap-Up: The 4 Pillars of Trusted Supplier Risk Intelligence
- 21:41Final Takeaways & Closing Thoughts
related talks
44mWhat Is a RAG System and How to Build One
01h 28mMastering RAG Evaluation with MLflow & Open Source LLMs
28mCombining Third-Party Data, Entity Resolution, and Graph Analytics for ESG Risk Mgmt
17mScaling Enterprise-Grade RAG: Lessons from Legal Frontier
23mSynthetic data that meets enterprise compliance needs
44mRAG for a Medical Company: Technical and Product Challenges AI & Data
15mEvaluating RAG Model Performance Metrics, Bias, and Interpretability
29mRAG for Data Engineers: Why Your Metadata Is the Missing Piece