Agentic RAG with DeepSeek-R1 — an intensive practical course on creating an agent-based Retrieval-Augmented Generation system using the local LLM DeepSeek-R1. You will step-by-step build a working application where agents, tools, tasks, and RAG components interact in a unified workflow for information retrieval, reasoning, and response generation.
The course is practice-oriented: instead of abstract theory, you set up a local environment, launch DeepSeek-R1 via Ollama, connect external data sources, design agent roles, and integrate everything into a complete Agentic RAG application.
About this course
Traditional RAG usually works on a linear scheme: a user asks a question, the system retrieves relevant documents, and the language model generates a response based on the found context. Agentic RAG extends this approach: an agent can independently choose tools, refine the task, manage searches, perform intermediate actions, and decide which data to use.
In the course, you will learn how to build such an architecture in practice: from a local model launch to setting up agents, tasks, tools, and the final application. Special attention is paid to working with DeepSeek-R1, local launches via Ollama, and designing the agent-based workflow.
What you will learn
- Understand the difference between traditional RAG and Agentic RAG.
- Launch DeepSeek-R1 locally using Ollama.
- Configure the environment for developing Agentic RAG applications.
- Create tools for information retrieval and performing external actions.
- Design agents with clear roles and responsibilities.
- Define tasks for agents and coordinate their execution.
- Build a comprehensive agent-based workflow for handling user requests.
- Integrate agents, tools, tasks, and RAG components into a unified system.
- Launch, test, and refine a ready-made application.
Course program
1. Introduction to Agentic RAG
You will be introduced to the key principles of Retrieval-Augmented Generation and understand why the agent-based approach is becoming important for complex LLM applications. This section explains when traditional RAG is sufficient and when it's better to use agents and tools.
2. Local launch of DeepSeek-R1 via Ollama
You will set up a working environment, install necessary dependencies, and launch DeepSeek-R1 locally. This approach allows experimentation with LLM applications without constant reliance on external APIs.
3. Creating tools for agents
You will implement tools that an agent can use for data retrieval, accessing external sources, and performing specific actions. This stage helps you understand how LLM accesses relevant information beyond its own context.
4. Designing agents and tasks
You will define agent roles, their responsibilities, and interaction logic. Instead of one universal component, the system will consist of specialized participants, each responsible for a particular part of the process.
5. Building the Agentic RAG workflow
You will integrate the model, tools, agents, tasks, and RAG logic into a single workflow. At this stage, the application will start executing a complete cycle: receiving a request, searching for data, analyzing the context, and generating a response.
6. Testing the finished application
In the final part, you will launch the finished Agentic RAG system, test its functionality on practical requests, and learn how to improve the quality of responses, agent structures, and tool usage.
Who this course is for
- For Python developers who want to create LLM applications using RAG and agents.
- For ML and AI engineers exploring the practical application of local language models.
- For specialists who want to understand how to use DeepSeek-R1 in real projects.
- For developers interested in Ollama, local LLMs, vector search, and agent-based architectures.
- For those already familiar with basic RAG and looking to transition to more flexible Agentic RAG systems.
What you will create as a result
By the end of the course, you will have a working Agentic RAG system based on DeepSeek-R1. You will understand how interaction between agents, tools, tasks, and information retrieval mechanisms is organized, as well as gain practical experience in developing agent-based LLM applications.
The completed project can be used as a basis for your own solutions: intelligent document search, AI assistant, internal knowledge base, research assistant, or a prototype of a corporate RAG application.
Required knowledge
- Basic understanding of Python and working with dependencies.
- General knowledge of large language models and prompting.
- It is desirable to understand the fundamentals of RAG, but the key differences of Agentic RAG are explained in the course.
- Experience with Git and the command line will be useful for launching the project.
Conclusion
Agentic RAG with DeepSeek-R1 is a practical course for those who want not only to understand the concept of agent-based RAG but also to build their own application with local LLM, tools, agents, and managed workflow. After completing the course, you will be able to design and launch more flexible RAG systems better suited for complex user requests and working with external knowledge.
