Building Agentic AI Systems: From Transactions To Conversations
Talkat Open Source Summit Europe 2025 · on Tuesday, 2025-08-26· at 11:55· Elicium 1 (Elicium Level 1)·
summary
Argues that production agentic systems require concrete engineering: a fine‑tuning pipeline, heavy GPU inference with caching, and an agent runtime with durable execution and memory, not just bigger LLM weights. Describes a reasoning loop—perceive, plan, observe/evaluate, act—plus multi‑agent orchestration, long‑running workflows, and evaluation strategies. Cites open source tooling such as vLLM, LangGraph, AutoGen and emphasizes three stacks—training, inference, runtime and durable execution and memory as the components that make agents work.
AI-written summary · about
chapters (9)
- 0:00Agentic AI systems intro
- 3:14SAS vs agentic systems
- 8:54Reasoning loops in design
- 17:24Three stacks for agentic systems
- 26:30Runtime stack for workflows
- 30:05Multi-agent architectures
- 35:00Trust in AI deployments
- 39:14Key takeaways
- 40:16Q&A on temperature
related talks
22mHOW TO BUILD YOUR OWN AI AGENT: THE COMPLETE ARCHITECTURE FOR ENGINEERING TEAMS
31mAgentic AI architecture patterns
28mAgentic AI
29mAgentic AI: What It Is and Why It Matters
22mAgentic AI for Engineering Leaders | Strategy & Execution Guide
16mBuilding Applications with AI Agents
34mAI Agents: Transforming Autonomous Decision-Making
40mBuilding Effective AI Agents Simple, Composable Patterns