In this module, the focus shifts from connecting to AI to actually building with it. Having learned how to call LLMs through code, you'll now apply those skills to real working applications that solve practical problems. The first project is Excel GPT, where AI translates plain English into spreadsheet formulas and VBA macros—turning natural language into functional automation. The second is a Bitcoin Analyzer that pulls real crypto market data and generates investment-style reports, giving you a taste of how AI can be paired with live data sources to produce meaningful analysis. Together, these projects show how quickly LLMs move from novelty to utility, and they set the stage for the more advanced pipelines and agentic systems coming later in the course.

In this module, you'll learn how to take real control over what AI produces—from the words it generates, to the format it returns, to the actions it can trigger. We start with prompt engineering: how to write prompts that reliably get exactly the response you need, including zero-shot, few-shot, and chain-of-thought strategies. From there, we move into structured output, where you'll see how to force any LLM to return clean JSON and use Pydantic to validate it for type-safe, production-ready results. Finally, we'll cover function calling—the mechanism that lets a model decide when to take action and invoke external tools. Together, these techniques form the complete pipeline for turning unpredictable text generation into dependable, controllable AI behavior.

This module moves beyond single API calls into the world of AI pipelines, where real applications come to life. Having mastered API calls, tools, and prompting in earlier modules, you'll now learn to chain these capabilities together into multi-step workflows using LangChain. We'll cover the four core building blocks—LLMs, Prompt Templates, Memory, and Chains—and show how they combine through LangChain Expression Language (LCEL) to create robust pipelines. You'll also see these ideas in action through the Beer Game, a classic supply chain simulation reimagined with AI agents making the decisions. By the end, you'll understand why production AI systems are rarely a single call and how to design the pipelines that power them.

In this module, we tackle one of the most important problems in applied AI: hallucination. LLMs are powerful, but on their own they confidently invent facts because they only know what was in their training data. Retrieval-Augmented Generation (RAG) solves this by grounding the model in your actual data. You'll learn why hallucinations happen and how RAG addresses them, then dig into the mechanics that make it work—how embeddings turn meaning into numbers, and how vector stores use those numbers to enable semantic search. From there, we'll walk through the full RAG pipeline end to end: chunking documents, embedding them, retrieving the right context, and generating grounded responses. By the end, you'll understand how to build AI systems that answer from your data rather than guessing.

In this module, we make the leap from chains to agents—the shift from AI that follows a fixed script to AI that reasons, plans, and decides its own next move. You'll start by understanding the fundamental difference between chains and agents, and when each is the right tool for the job. From there, we dig into the ReAct pattern (Reason, Act, Observe, Repeat), which is the core loop that gives agents their problem-solving ability, and you'll see how function calling in a loop lets them use tools to interact with the world. Finally, we'll bring it all together with LangGraph, the framework for building stateful, multi-step agent workflows that can handle real complexity. By the end, you'll be able to design AI systems that don't just respond—they think and act.

In this module, you put it all together. Over a full week, you'll work in teams to design, build, and ship a real AI project using the techniques you've practiced across the course. By this point, you've completed more than a dozen labs across API calls, prompt engineering, structured outputs, LangChain pipelines, RAG, and agents—this hackathon is where those individual skills converge into something whole. There are no new lectures and no new labs. Instead, this is a sprint: Pick a problem worth solving, choose the right architecture, build a working demo, and present it. The best projects are not the most complex—they're the most focused. By the end of the week, you'll have a portfolio-ready artifact that shows what you can build with the same building blocks behind today's leading AI products.