Build an Agentic AI System From Scratch — Production-Grade, Not a Tutorial Demo

The right starting point is the architecture, not the code. Before writing any Python, the workshop covers the Glass-Box Context Engine design — what agents you need, what context each one manages, how MCP connects them, and how the RAG pipeline integrates. This upfront architectural thinking is what separates agentic AI systems that scale from those that break under complexity.

Initially yes — frameworks provide shortcuts that speed up the first prototype. But building from scratch gives you a system you fully understand, can debug at any layer, and can extend without fighting framework limitations. The patterns you build in this workshop become your own reusable framework that is better suited to your specific use case than any general-purpose library.

You build: the semantic blueprint generator, the MCP orchestration layer for agent coordination, individual specialized agents with their own context management, the high-fidelity RAG pipeline with citation tracking, the Glass-Box observability layer, the safeguard components for input and output validation, and the production deployment configuration. Every component is implemented in Python during the live 6-hour session.

With the context engineering architecture from this workshop, building your first production-quality agentic AI system takes a few days of focused development after completing the workshop. The 6-hour live session gives you the complete architecture and working code as a foundation. Subsequent systems built on the same patterns are faster because the core components are reusable.

This workshop builds from scratch to give you deep architectural understanding. In practice, many developers build the core context engineering layer from scratch while using libraries for specific components (embedding, LLM clients, vector stores). The workshop teaches you which components benefit from custom implementation and which are fine to use off the shelf.

Memory engineering is typically the most complex part — designing how context is stored, retrieved, and expired across the working, episodic, and semantic memory layers. The workshop dedicates significant time to memory engineering because getting it right is what separates agentic AI systems that stay reliable over long conversations from those that degrade. Denis Rothman draws on his decades of AI memory system experience for this module.