Grounded Reasoning Systems for Cloud Architecture
summary
Built an AI copilot that uses LangGraph, Flyte and AWS Bedrock (Claude 3) to orchestrate many specialized agents and a domain knowledge graph as long‑term memory. Production system generates architecture recommendations, clones staff agents for parallel design proposals, and uses an internal human eval tool "eagle eye" for scoring. Semantic vector search helps retrieval but fails to scale on graph data, so graph‑based search and structured outputs improve continuity and debuggability; OpenTelemetry is planned for observability.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 0:18Why now
- 1:31AI Meets Architecture
- 2:49Recommendations
- 3:53Challenges
- 7:15Architecture Retrieval
- 8:28Knowledge Graph
- 10:05Complex Reasoning
- 14:07Control Flow
- 15:12Agents
- 15:54Workflow
- 18:49Evaluation
- 21:39Evaluation Tool
- 22:02hallucination
- 22:54conclusion
related talks
31mAI Agent Orchestration for Enterprise Arch with AWS Bedrock, Flyte & LangGraph
01h 26mThinking in Graphs: The Cognitive Foundations of Agentic AI
39mBuilding Safe AI Agents
24mBuilding GenAI applications with Amazon Bedrock
43mArchitecting the Future: Modern Solution Design with Agentic AI
30mBeyond Monolithic AI: Cloud Native Patterns for Dynami...
29mBuild an enterprise Agentic AI platform at Powercor
48mHow to Make Smart Architecture Decisions when Building Gen AI Apps