Building LLM Agents Made Simple a
summary
A hands‑on build of a back‑office automation system using LlamaBot inside Marimo notebooks maps workflows into three concrete agents: receipt processor that OCRs PDFs and structures fields, invoice writer that generates templated invoices, and a coordinator that composes them. Setup uses Pixi and Ollama models; code runs in the cloud so notebooks execute live. Demonstrations produce working code but show extraction errors (wrong total), highlight the need for structured‑generation schemas and rigorous evals to catch model drift.
AI-written summary · about
chapters (12)
- 0:00Building LLM agents
- 1:08Tutorial setup and cloud walkthrough
- 5:26Speaker and project motivation
- 9:18Agents versus workflows
- 15:57Traditional automation first
- 20:39Receipt processor workflow
- 31:07Structured data with Pydantic
- 43:18Wrapping workflows as tools
- 59:17Invoice writer agent
- 1:02:34Agent orchestration with chat
- 1:14:56Key takeaways and framework
- 1:17:49Q&A on LlamaBot and evaluation
related talks
41mBuilding LLM Agents – Challenges and Learnings
01h 43mBuilding LLM Agents (Hackathon)
35mBuilding LLM-based Agents
19mLLM codegen fails and how to stop 'em
26mFrom APIs to Agents: Enterprise AI using Java, Ollama & MCP
50mIntelligent Agents, the New Frontier of LLMs
47mOwn Your LLM: Running Generative and Agentic AI Locally
09mBuilding LLM Agents with Tool Use