Beyond vector search: knowledge graphs for an AI-native SDLC
summary
Argues that running agents at organizational scale requires a schema-driven prod mode memory rather than raw vectors, and shows a real build: a knowledge graph exposed as an MCP tool that ingests metrics (and traces converted to metrics), runs hourly pipelines, and populates nodes via drain templating plus fuzzy matching (Jaccard-style scoring) instead of another LLM. Logs are hardest; templates are filtered by schema. A prod incident with a 5x traffic surge shows the knowledge graph acting as guardrails to find the real root cause.
AI-written summary · about
chapters (15)
- 0:00AI-native SDLC intro
- 3:35AI agents evolving dev
- 5:00Prod vs dev challenge
- 7:10Knowledge graphs for observability
- 8:33Schema for knowledge graph
- 10:21Metrics to graph mapping
- 13:08Ambiguity in metric matching
- 15:55Live data population
- 17:49Traces to graph edges
- 18:53Logs to knowledge processing
- 21:43Real-time population strategies
- 22:46Graph visualization
- 25:19MCP server access
- 26:24AI agent error analysis
- 28:53Key takeaways & future
related talks
12mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
21mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
30mAgentic RAG with Knowledge Graphs Explained
28mData-Empowered AI Assistants
21mFrom Data to Knowledge to Action - The Graph Intelligence Platform
15mMultiple Knowledge Graphs for Sophisticated RAG
29mAgentic AI with Knowledge Graphs on the JVM: A Music Graph in Practice
20mEnhancing RAG with Multi-Agent Integration