Building Your AI Assistant: From First Principles to Agentic With Ivan (ex-Manus.ai)
Join us at SQ Collective for beginner friendly workshop led by Ivan Leo (Previously AI Engineer @ Manus & Maintainer of Instructor).
Ivan will guide you on how you can build the core architecture of world-class agents like OpenClaw and Pi, starting from a blank Python file from first principles.
He will run through the concepts of building agents at a high level while showing demos of how it is done in real time. All code and slides will be provided after the session for you to replicate.
This session is structured around the two foundational pillars of Ivan’s "OpenClaw from Scratch" series.
Part 1: The Lay of the Land (The Architecture)
Before we code, we must understand the "why." We’ll break down why agents like OpenClaw and Pi are effective and how to architect them without the bloat of heavy frameworks.
The Anatomy of an Agent: Understanding the loop between LLMs, context, and capabilities.
Gemini-Powered Logic: Why we’re pushing gemini-3-flash to its limits for tool calling.
Memory & State: Implementing persistent memory so your agent doesn't forget who you are.
Lifecycle Hooks: How to build "guardrails" and approval queues into the agent’s life cycle.
Part 2: It’s Alive! (Self-Extending Capabilities)
The "Frankenstein Moment." This is where the agent stops being a chatbot and starts being an operator.
The Tool Factory: Using Pydantic to build a runtime that makes defining new tools (Bash, Edit, Write) effortless.
Self-Extension: The "black magic" of agents—teaching your agent to write its own tools, modify its own source code, and hot-reload its capabilities on the fly.
Sandboxed Execution: A look at running code safely so your agent can experiment without breaking your machine.
Who Is This For?
AI Builders & Engineers who want to go beyond simple RAG and into agentic workflows, or just wants to understand agent building from first principles.
Founders & Operators looking to understand the how-to of AI-native automation.
The Curious beginner who want to see the cutting edge of what's possible with models that act. (some understanding of code is still required)
About Your Instructor
Ivan Leo was previously an AI Engineer at Manus and a core maintainer of Instructor, an open-source library used by millions to get structured data from LLMs. He is a fixture in the Singapore AI scene, known for turning complex research into production-grade systems.
Event Details
Date: Thursday, 12 March, 6:30–9:30pm
Capacity: Limited to 60 pax
Location: SQ Collective, 65 Mohamed Sultan Rd, Singapore
Note: Bring your laptop. This is a hands-on session.
First 20 minutes will be a walk through of high level concepts, and the remainder will be demo, hands-on and QnA.