The need
Engineering students hear about AI agents constantly but rarely build one. A separate problem made a good exercise: students struggle to navigate long PDFs and scattered Drive folders, in several languages.
What I owned
The whole session: the program, the slides, the handouts, the two workflows, the system prompts and the live delivery.
The solution
Participants built an AI study assistant available on Telegram, specified like a real client project, with a brief, constraints and guardrails.
Key modules
Foundations
What an agent is, how it differs from a chat assistant, and where agents are used across industries.
Workflows and APIs
Automations as graphs, how data changes as it moves through them, node types in n8n, and how API requests connect outside services.
RAG
Why models have blind spots with unseen data, how chunking and embeddings work, and how retrieval fixes this.
Prompting and guardrails
System prompt best practices, with an assistant that never invents information and says clearly when content is missing.
How it was built
The plan moved from concepts to practice: agents, then workflows, then n8n and APIs, then the project brief, RAG, the two workflows, prompting and testing. Each participant received reference handouts on APIs, agentic systems, deep learning, transformers, RAG and system prompts.
Results
Participants left with the understanding of how agents work and a retrieval-powered assistant they had built themselves, from theory to a working system in one session.