Give Your LLMs a Left Brain
summary
Shows that combining Backstage and Red Hat OpenShift Container Platform with a knowledge-graph layer fixes LLM weaknesses: a retrieval flow queries the graph first, falls back to a vector DB, then passes a subgraph as context to the LLM. Claims include deterministic, explainable answers versus vague vector-only outputs; a real example ingests 1.5 million documents (Panama Papers) and surfaces hidden relationships; a customer survey finds 80% of respondents already deploying LLM apps. The architecture proves more accurate for reasoning, auditing, and role-based data access.
AI-written summary · about
chapters (11)
- 0:00Introduction to LLMs
- 3:36LLM use cases in business
- 6:50Limitations of LLMs
- 7:11LLM reasoning gaps
- 17:40How LLMs work under the hood
- 24:44Context window limitations
- 25:05Combining LLMs with knowledge graphs
- 33:11Risks of unchecked LLM outputs
- 38:09Architecture for LLM integration
- 44:16Future of responsible AI
- 45:41Resources and closing
