Agentic AI & GenAILightning talk20min
Build your own at home Moroccan Harness
A live, no-framework demo of building an AI agent harness from scratch in about 200 lines of code. The talk explains loops, tools, context, and permissions, using Moroccan examples like Darija chat, tagine shopping, and local task automation to demystify Claude Code or Cursor.
talk.summaryAiDisclaimer
Abderrahman JAIZERegulaitor
Everyone's talking about AI agents, but most people treat the "harness" (the loop that makes an LLM actually do things) as a black box. In this talk I build one from scratch, live. Starting from a bare model call, we add a tool loop, file access, and a permission layer until we have a working coding-and-task agent, no framework, no magic. Along the way I ground every concept in home-grown Moroccan examples: an agent that plans a tagine shopping run, answers in Darija, and automates a real local task. You'll leave understanding exactly what's inside Claude Code or Cursor, and with a repo you can run at home tonight.
What a "harness" is (the spine of the talk): an agent harness is everything wrapped around an LLM that turns it from a chatbot into an agent, the loop, the tools, the context management, the permissions. Claude Code, Cursor, etc. are harnesses. Your talk's thesis: this is not rocket science, you can build a working one at home in a couple hundred lines, and we'll do it live with Moroccan examples.
Tagline: From LLM to agent in 200 lines, dial l'agent b yeddek (build the agent with your own hands)
What a "harness" is (the spine of the talk): an agent harness is everything wrapped around an LLM that turns it from a chatbot into an agent, the loop, the tools, the context management, the permissions. Claude Code, Cursor, etc. are harnesses. Your talk's thesis: this is not rocket science, you can build a working one at home in a couple hundred lines, and we'll do it live with Moroccan examples.
Tagline: From LLM to agent in 200 lines, dial l'agent b yeddek (build the agent with your own hands)
Abderrahman JAIZE
I'm a Senior Data Scientist based in Brussels with around 8 years of experience taking AI systems from a business problem all the way through data, modeling, deployment, and monitoring. These days I focus on Generative AI: building multilingual RAG chatbots, fine-tuning large language models like GPT-4, Claude, and Gemini, and designing agentic AI workflows with LangGraph, AutoGen, and CrewAI. I currently work with Zazmic, a Google Partner, where I've shipped high-volume AI solutions serving more than 10M requests a month on GCP Vertex AI.