Harrison Chase teaches how to build apps where users chat with their own data. The course covers loading documents, splitting them, using vector stores and embeddings, and building retrieval systems that work within LLM context windows.
Hi. I'm thrilled to be back with Harrison Chase, creator of LangChain and instructor for this new short course LangChain Chat with your Data. I'm happy to be back and excited to get into these features and use cases for everyone creating innovative things using LangChain So if you're just joining us, we previously released a course on LangChain titled LangChain for LLM Application Development, which if you haven't taken yet, is a good place to learn the basics of LangChain. This course goes further and teaches you one of the most powerful and useful things you can do with LangChain, which is to build an application to chat with your data. Let's say you have an employee handbook or have information in a vast library of data and you need to find relevant information quickly. What if you could just ask a chatbot exactly what you need and and it could provide it for you? That's what we're covering in this course with a variety of examples and techniques. What I'm seeing across a lot of large companies is that chatting with their data is one of the most commonly suggested internal project ideas, and not just for businesses, but individuals too. For example, being able to ask questions about the content of a specific white paper or research paper without you needing to reread the whole thing. That helps you get the information you need more quickly. Here are the lessons that make up this course in chatting with your data. First, we'll cover document loading. LangChain currently provides over 80 unique loaders to handle accessing various data sources, including audio and video. You'll learn the basics of data loading and try several examples. Next is document splitting. This is seemingly straightforward, but splitting data has many subtle considerations that influence how effective vector stores are when retrieving information. We'll get hands on and learn important considerations when selecting a splitter. We'll then cover vector stores, which I just mentioned. You'll review the concept of embeddings in vector store integrations in LangChain and use practical examples of these technologies. After that is retrieval. So how do you access and index the data into a vector store and go beyond semantic queries so that you can retrieve the most relevant information? Then question answering we'll build a one pass question answer solution and explore some of the important considerations when working with the limited context window of most LLMs. And finally chat. We'll showcase a chat over your documents application. LangChain supports methods that help you track and select relevant information from both the conversation and data sources, and you'll build your own chatbot using using LangChain. Thanks, Harrison. And so that's a lot of very exciting and I think important topics for building applications using LLMs and for chatting with your data. And so I hope you enjoyed this short course as well as the earlier one that Harrison and I had taught covering the basics of LangChain. And after this course, I hope you also go build some really cool applications to let yourself or others chat with data.
