Agents Everywhere
summary
Claims that complex multi-agent LLM systems fail up to 75% of the time lead to a defensive design: a stateful directed cyclic graph engine called Langraph DCG that can rollback, checkpoint, and constrain LLM nondeterminism. Demo code shows linear hello-world graphs, branching routers, loops, tool calls (web weather) and a Langraph+Redis memory stack: short/long/episodic memory, a memory server, a Hugging Face fine-tuned model for semantic caching and a dock-to-cache pipeline that cuts repeated LLM calls (≈30% of queries) and lowers latency and cost.
AI-written summary · about
chapters (8)
- 0:00Introduction and agenda
- 1:05What are agentic systems
- 10:21Core agent properties
- 24:18Agentic architectures and patterns
- 33:58LangGraph deep dive
- 39:19Memory in agentic systems
- 51:04LangGraph demos
- 1:00:59Emerging trends and closing
related talks
01h 59mBuilding Agentic and Multi-Agent Systems with LangGraph
46mDemystifying AI Agents
56mHerding LLMs: Predictable Outcomes from Unpredictable Agents
49mPushing LLMs over the Edge: Exploring the Limits of the Possible
29mBuilding Modular AI Agents with LangGraph, MCP, and
28mAI Agents Graph: Your Following Tool in Your Java AI Journey
01h 43mBuilding LLM Agents (Hackathon)
41mEnhancing LLMs with Graph Technology