From LLM basics to agentic AI systems
A structured, step-by-step journey for software engineers — moving from fundamentals all the way to building production-grade, agentic AI applications.
Choose the track that matches your level
Jump straight into the phases that matter most for you, or follow the full sequence from foundations to capstone projects.
Foundation Track
Start with LLM basics, prompting, and AI-assisted engineering workflows.
Builder Track
Move into APIs, security, local models, context engineering, and RAG.
Architect Track
Design agentic workflows and finish with portfolio-ready projects.
Build a solid foundation before touching any AI tooling. Everything advanced rests on what you learn here.
- Explain in plain language how an LLM turns a prompt into a response.
- Write structured prompts using at least five distinct techniques.
- Choose an appropriate model for a given task, cost, and privacy constraint.
Learn what Large Language Models are and how they work. Grasping these concepts lets you write better prompts and control costs.
Prompts are your primary interface with AI. Master these five techniques for dramatically better results.
Know the types of models available so you can pick the right one for any task.
Start using AI tools to enhance your daily engineering workflow — before you write a single line of AI code yourself.
- Use an AI coding assistant fluidly inside a real project.
- Know the practical limits of today's coding agents.
- Build one automated workflow that includes an AI step.
Get hands-on with the everyday tools before going deeper.
Learn how AI fits into automated, multi-step pipelines.
Move from chat interfaces to programmatic access. Learn to call AI models like any other API — safely.
- Call an LLM API from the terminal and from code.
- Keep API keys and rate limits secure by default.
- Route requests across models to balance cost and capability.
The same request, two ways: raw HTTP and an SDK.
Treat AI APIs like any other production dependency — cost and abuse matter.
One endpoint, many models — choose per request.
Run open-source models on your own hardware. Gain privacy, reduce costs, and understand what's under the hood.
- Run an open-weight model locally and chat with it offline.
- Explain weights and quantization in your own words.
- Know when local models beat hosted APIs (and when they don't).
The local-model ecosystem you'll actually use.
Understand what's actually running on your machine.
Context is everything when working with large codebases (50,000+ lines). This phase teaches you how to manage what the AI knows and when.
- Feed only the most relevant context to a model.
- Distinguish short-term, long-term, and semantic memory.
- Connect a model to a tool using MCP.
More context isn't better — the right context is.
Agents need different kinds of memory for different jobs.
The open standard for connecting models to tools and data — “USB-C for AI”.
Retrieval Augmented Generation (RAG) lets your AI work with private, organizational data that was never in its training set.
- Explain embeddings and similarity search.
- Stand up a vector database and ingest documents.
- Wire retrieval into an LLM query end-to-end.
RAG stands on three ideas. Get these before building.
Put the pieces together and test on real data.
Go beyond single-model calls. Learn to orchestrate multiple specialized agents that collaborate to solve complex problems.
- Describe the roles agents play in a multi-agent system.
- Implement hand-offs between agents.
- Build and run a working multi-agent pipeline.
Start small — two cooperating agents beat one giant prompt.
Pick one framework and build something real with it.
Reading and exercises are not enough. You cement everything by building real, integrated projects from scratch.
- Integrate multiple techniques (MCP, RAG, agents) into one project.
- Work fluently inside an AI-assisted IDE.
- Produce documented, portfolio-ready work.
Projects are where the learning becomes permanent.
Pick one that excites you and ship it.