Agentic AI Nanodegree Projects
This is a collection of all the projects I built while completing Udacity's Agentic AI Nanodegree. The program was structured around four main courses, each going deeper into how large language models can be used as the backbone of autonomous agents.
The first course focused on the foundations of LLM reasoning. I worked through understanding how models like GPT handle different prompting strategies, from basic zero shot and few shot techniques to more structured approaches like chain of thought. The key project here involved building a system that could break down complex questions into smaller reasoning steps. What surprised me was how much the output quality depended on the structure of the prompt rather than the raw capability of the model.
The second course moved into building actual agents. This is where things got really hands on. I built agents that could use tools, call APIs, and make decisions about which actions to take based on the current context. The ReAct (Reasoning + Acting) pattern was central to this course, and implementing it from scratch gave me a much better understanding of how frameworks like LangChain abstract these patterns. I built a travel planning agent here that could search for flights, check weather data, and put together itineraries based on user preferences.
Course three covered agentic workflows using LangGraph. LangGraph lets you define agent behavior as a graph with nodes and edges, which makes it possible to build more complex flows with branching, loops, and conditional logic. I built an AI project manager that could break down project requirements, assign tasks, and track progress. The graph based approach was a shift from the linear chains I had been building, but it opened up a lot of flexibility for handling real world scenarios where the next step depends on what happened before.
The final course brought everything together with multi agent systems. Instead of a single agent doing everything, I built systems where multiple specialized agents collaborate to solve problems. The capstone project was an automated sales system with separate agents handling lead qualification, product recommendations, and follow up communications. Each agent had its own tools and context, and they communicated through a shared orchestration layer.
Looking back, the biggest takeaway from this nanodegree was understanding the spectrum from simple prompt engineering to full multi agent orchestration. Each level adds complexity but also opens up new capabilities. The projects are all in the repository with documentation and instructions for running them locally.