Aleksey Seleznev releases a free R video course on building AI tools

Aleksey Seleznev has published a free R video course on building AI tools, chat interfaces and multi-agent systems. It is laid out as a practical route for people who already write R and want to build LLMs into their workflows, rather than just use web chats.

What the course covers

The main idea is to show that the R ecosystem now works not only for classic analytics and visualization but also for full AI application development. The author does not offer a set of scattered tricks. He builds on a connected stack: `ellmer` for working with models, `mcptools` for MCP, `ragnar` for RAG, `shinychat` and `querychat` for interfaces, and `mini007` for agent scenarios. As a result, the course reads less like a survey of new releases and more like an attempt to give R developers a ready-made path into applied AI.

"This course is not about abstract examples and toy demos."

The materials are also free, in the form of an online book with video lectures, notes and code examples. That lowers the barrier for people who have worked in R for years but still saw modern AI frameworks as Python territory. The course does two things at once: it gives a learning structure and shows which packages to use right now, without assembling a stack yourself from dozens of repositories and articles.

Seven practical modules

The program consists of seven sequential lectures. First, the author covers basic work with LLMs through the `ellmer` package: setting up API keys, creating chats, extracting structured data from text and building a simple interface. The course then moves to more applied topics: moving the model into a Telegram bot, managing user context and saving dialogue history between sessions.

• Connecting different LLM providers directly from R
• Building AI chats and web interfaces on Shiny
• Running an MCP server and an MCP client to work with external tools
• Building a RAG system with embeddings, DuckDB and hybrid search
• Developing multi-agent scenarios that generate and run R code

In the second half, the focus shifts to more complex architectures. One module covers MCP as a way to connect models to data and functions, including connecting to tools such as Claude Desktop. Another covers the RAG approach: embeddings, a DuckDB vector store and answers based on your own documentation. Separate sessions cover configuring the `shinychat` interface, how `querychat` works (it translates natural-language requests into SQL), and multi-agent systems where the AI not only answers but also coordinates actions, evaluates the quality of results and executes code.

Who the course is for

According to the author, the course targets data analysts, R developers and anyone who already writes R code with confidence. It is not an introduction for absolute beginners: without basic language syntax and an understanding of how packages work, the pace will most likely be too fast. For practitioners who want to move quickly from "I tried a chatbot" to a working prototype, though, the program looks quite practical.

The course's strength is that it covers several working scenarios that people usually learn piecemeal. LLM APIs, bots, web interfaces, data access via MCP, search over your own knowledge base and agent patterns are all collected in one place. For corporate teams this is especially useful: an analyst or BI developer can keep the main language and stack and add AI features directly to existing R projects, dashboards and internal services. The free format also matters: a team can use the course as an internal entry point without a separate training budget.

What this means

R remains a niche language compared with Python in AI development, but courses like this show the gap is closing quickly at the level of applied tools. For the Russian-speaking community, it is a ready, structured way into LLMs, MCP, RAG and agent systems without leaving the familiar ecosystem.