In this article, we will talk about LLM Zoomcamp, our free online course to get started with real-life applications of LLMs. In 10 weeks, you will learn how to build an AI system that answers questions about your knowledge base.
We will cover different aspects of this course so you can learn more about it:
Who is the course for?
Before we get into the details, it’s important to know what skills you should have to comfortably join the course.
Here are the main prerequisites for the course:
- Comfortable with programming and Python
- Comfortable with command line
- Docker
- No previous exposure to AI or ML is required
Course Curriculum
- Module 1: Introduction to LLMs and RAG
- Module 2: Open-source LLMs
- Module 3: Vector Databases
- Module 4: Evaluation and Monitoring
- Module 5: LLM Orchestration and Ingestion
- Module 6: Best Practices
- Module 7: Bonus: End-to-End Project Example
Let’s quickly review each module, focusing on the main points and the tech you’ll use.
Module 1: Introduction to LLMs and RAG
We introduce the core ideas behind Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG). You’ll set up your development environment, learn how retrieval works, and start experimenting with APIs and search tools. By the end of this module, you’ll have a basic RAG setup and be familiar with text search fundamentals.
You will learn to:
- Set up your environment for LLM and RAG experimentation
- Understand the basics of retrieval and search
- Use the OpenAI API to integrate LLM capabilities
- Build a simple RAG system
- Implement basic text search with Elasticsearch
Module 2: Open-source LLMs
We dive into the world of open-source LLMs, providing hands-on experience with popular, freely available models. You’ll learn how to configure a GPU environment, access models from the Hugging Face Hub, and even run LLMs on CPUs when GPUs aren’t available. This module ends with creating a simple UI to see your model in action.
You will learn to:
- Set up and optimize a GPU environment
- Access and use open-source models from Hugging Face
- Run models on a CPU using Ollama when GPUs aren’t available
- Create a basic, interactive UI with Streamlit for testing your model
Module 3: Vector Databases
This module covers how to use vector databases for effective search and retrieval. You’ll learn to create embeddings (vector representations of text), index them, and use vector search to improve RAG performance.
You will learn to:
- Create and index embeddings for vector-based retrieval
- Implement vector search using Elasticsearch
- Conduct offline evaluations to assess your retrieval system
- Work hands-on with dlt to practice embedding indexing and search
Module 4: Evaluation and Monitoring
We focus on evaluating your RAG system and setting up monitoring tools. You’ll explore different metrics to judge your system’s performance and set up a feedback loop for continuous improvement. Grafana dashboards will help you visualize insights and track system usage.
You will learn to:
- Perform offline evaluations of your RAG pipeline
- Use Cosine Similarity and LLM-as-a-Judge metrics to assess retrieval
- Track chat history and collect user feedback for iterative improvement
- Build Grafana dashboards to monitor performance in real-time
Module 5: LLM Orchestration and Ingestion
This module teaches you how to efficiently manage data ingestion for LLMs.
You will learn to:
- Ingest data seamlessly
- Set up a smooth data pipeline for LLM projects
- Prepare data for scalable and efficient processing in RAG systems
Module 6: Best Practices
We dive into advanced techniques for refining your RAG pipeline, from improving retrieval quality to enhancing search relevance. You’ll practice hybrid search methods, document reranking, and explore using LangChain for more complex applications.
You will learn to:
- Apply best practices to optimize your RAG pipeline
- Use hybrid search techniques to increase retrieval accuracy
- Implement document reranking to enhance search results
- Set up a hybrid search with LangChain for advanced retrieval tasks
Module 7: End-to-End Project
You’ll bring everything together in a practical project. You’ll apply all the skills you’ve learned to complete an end-to-end project, from data preprocessing to deploying your solution.
You will learn to:
- Build an end-to-end project using RAG techniques
- Practice preprocessing text data for specific use cases
- Apply learned techniques in a real-world project
The course description on GitHub provides a detailed overview of the topics covered each module. To dive into the content, you can see the video lectures, slides, code, and community notes for each course module.
If you’re ready to join the next cohort of the course, submit this form to register and stay updated.
Theory and practice
We make LLMs theory accessible and engaging through real-world examples. We also demonstrate code directly in the lectures to show the implementation of concepts. This way, you can easily apply them in your projects.
For instance, in one of the lectures about a linear algebra refresher, the lecturer switches between screens. Firstly, they explain the concept of the dot product of two vectors, and then they demonstrate its implementation using Python.
Course assignments and scoring
Homework and getting feedback
To reinforce your learning, we offer regular homework assignments. Your scores are added to a leaderboard, creating friendly competition among course members and motivating you to do your best.
For support, we have an FAQ section with quick answers to common questions. If you need more help, our Slack community is always available for technical questions, clarifications, or guidance. Additionally, we host live Q&A sessions called “office hours” where you can interact with instructors and get immediate answers to your questions.
Learning in public approach
A unique feature is our “learning in public” approach, inspired by Shawn @swyx Wang’s article. We believe everyone has something valuable to contribute, regardless of their expertise level.
Throughout the course, we actively encourage and incentivize learning in public. By sharing your progress, insights, and projects online, you earn additional points for your homework and projects.
Sharing your work online also helps you get noticed by social media algorithms, reaching a broader audience and creating opportunities to connect with individuals and organizations you may not have encountered otherwise.
Course projects for your portfolio
If you’ve ever participated in an interview or conducted online research, you likely understand the significance of personal projects. To receive a certificate, you’ll need to finalize and submit an end-to-end RAG application. It allows you to choose a problem that interests you, find a suitable dataset, and develop your model.
DataTalks.Club community
DataTalks.Club has a supportive community of like-minded individuals in our Slack. It is the perfect place to enhance your skills, deepen your knowledge, and connect with peers who share your passion. These connections can lead to lasting friendships, potential collaborations in future projects, and exciting career prospects.
Conclusion
LLM Zoomcamp is a structured and practical introduction to applying Large Language Models in real-world contexts. Over 10 weeks, you gain hands-on experience, from setting up retrieval systems to building a complete RAG application.
Each module is crafted to build useful skills step-by-step, ensuring you can put what you learn into practice. If you’re interested in learning about and applying LLMs, joining the next cohort is a good way to start.
Register for the next LLM Zoomcamp cohort and stay updated on start dates by filling out this form.