Ship production R + LLM
workflows that actually work
The bridge between tidymodels and modern LLMs — tool-calling with ellmer, RAG with ragnar, production deployment via Shiny and vetiver, rigorous evaluation with vitals.
Official Partner
Every enrollment includes 1 month of prebuilt VM access with R and every package pre-configured. Zero setup — sign up, start coding. Explore membership for annual VM access across our platform.
Most R users get stuck between
tidymodels and LLMs
You know how to fit a model. You’ve played with ChatGPT. But the bridge between them — tool-calling, RAG, production deployment, evaluation — is where everything breaks.
Python tutorials don’t translate
Every LLM course is in Python. You spend hours translating LangChain only to discover half of it doesn’t have an R equivalent.
Tool-calling feels like black magic
You read about "agents" and "tools" but can’t figure out how to register a tidymodels prediction function as something an LLM can actually call.
RAG tutorials skip the hard parts
They show chunking on a toy example. Then you try it on your PDFs and retrieval quality is 20%. Nobody teaches evaluation.
Deployment is an afterthought
Your LLM app works in RStudio. Behind a Shiny app that doesn’t leak tokens or crash? That’s where projects die.
R-native from end to end
ellmer, ragnar, querychat, mall, vitals, vetiver. The modern R AI stack taught by the people who use it in production.
Tool-calling that calls your real models
Register a tidymodels workflow as a tool. NL interfaces to ML pipelines. Users ask questions, get predictions, never leave R.
Production RAG with proper eval
Hybrid retrieval (BM25 + embeddings). Chunking strategies that work. RAG evaluation with vitals so you know when your system is good.
Full Shiny + vetiver deployment
Package models with vetiver, wrap in Shiny, monitor cost and latency, add observability. Your app in production, not your laptop.
Watch the modern R AI stack
write itself
Most courses teach packages in isolation. Here you learn how ellmer, ragnar, tidymodels, and Shiny compose into real systems.
The terminal on the right is typing real R code, live. Three examples cycle automatically — or click any tab to jump to it.
See your RAG pipeline come alive
Watch documents flow through chunking, embedding, hybrid retrieval, and context augmentation — into an LLM that answers with sources. This is Module 19 in one diagram.
Six portfolio-worthy systems
Not toy examples. Complete, deployable R applications that demonstrate production-grade ML + LLM integration. The kind recruiters actually want to see.
Full tidymodels ML pipeline
Recipes, workflow sets, regularized regression, random forests, neural nets, cross-validation, hyperparameter tuning.
NL → SQL database analyst
Natural language interface to a relational database. Tool-calling, safe query execution, structured output extraction.
Document Q&A system
Hybrid retrieval over multiple documents. BM25 plus embeddings, chunking strategies, retrieval tool registration.
LLM evaluation harness
Eval datasets, statistical model comparison, RAG-specific evaluation design. Know when your system is actually good.
Deployed ML model API
Package a tidymodels workflow with vetiver, version it, deploy as an API, monitor drift in production.
Production Shiny chatbot
Full LLM chatbot in Shiny. Tool-calling, RAG, prediction explanation, cost/latency monitoring, proper UX.
Built by R’s most respected
LLM practitioners
Two instructors. One deep in package development and production R. One at the intersection of modeling, deployment, and modern AI workflows. Both writing R in production every day.
Apache Arrow Core Maintainer
Independent R consultant with 15 years in R and 10+ years shipping R in production, plus a PhD in Statistics. Core maintainer of Apache Arrow for R and author of Scaling Up with R and Arrow (CRC Press, 2025). Deep experience across pharma, public health, academia, and startups.
Co-Creator of Tidy Finance & EconDataverse
Financial economist turned data scientist and product manager. Builds interactive dashboards, automated reporting, and custom R/Python packages. Extensive experience with CI/CD and Shiny deployments — Docker, ShinyProxy, Posit Connect, GCR.
21 modules. 110+ lessons.
One complete path.
From tidymodels foundations through production LLM apps. Click any module to see every lesson.
Build a rigorous ML foundation using tidymodels — from exploratory analysis through deployment. Sports data throughout. Real workflows, not toy examples.
You don’t need a separate ML course after this
Most LLM courses handwave the ML side. Here, you get the full tidymodels treatment — preprocessing with recipes, tuning with workflow_sets, deployment with vetiver. By the time you hit Phase II, you’ve built production ML pipelines. The LLM layer sits on top of real modeling foundations.
Build the core LLM skill stack in R using ellmer. Prompts, conversations, structured data extraction, multimodal input, and model selection including local models via Ollama.
Prompt engineering isn’t magic — it’s structured thinking
You’ll go from basic chat to few-shot prompting, structured output that returns clean data frames, and multimodal workflows that extract tables from images. This is the skill stack that turns LLMs into actual R tools instead of toys.
The core differentiator of this course. Register R functions as tools. Build RAG systems with proper retrieval evaluation. Run NLP pipelines with mall. Measure everything with vitals.
Build a working RAG pipeline and learn how to evaluate it
Eight lessons on RAG covering BM25, embeddings, hybrid retrieval, tool registration, and context augmentation. Then eight lessons introducing evaluation with vitals, including RAG-specific eval design. By the end, you’ll have a working RAG system and a principled starting framework for evaluating it — with clear pointers to the deeper evaluation literature so you can go further when you need to.
Put it all together. Build complete LLM applications with Shiny, querychat for NL-to-SQL, and shinychat for production chatbots. Tool-calling, RAG, model explanation — all wired into a real deployable app.
You leave with a deployable app — not a notebook
By Module 21, you’re wiring everything together. Tool-calling inside Shiny. RAG inside chatbots. NL-to-SQL with querychat. Model explanation in-app. This is the portfolio project recruiters actually want to see — not another Kaggle notebook.
Two ways to get in
Presale pricing ends May 7, 2026 at 11:59pm ET. Same course, later date, higher price.
- 21 modules, 110+ lessons
- Full tidymodels + LLM R stack
- Tool-calling, RAG, evaluation, deployment
- Production Shiny chatbot capstone
- Dr. Nic Crane + Dr. Christoph Scheuch
- Certificate of completion
- All presale content included
- Launch access at course release
- Same 1-month VM access
- Same lifetime access
Why pay more to wait?
AthlyticZ Members enroll at $639 · Explore Membership →
Quick answers
Build the R + LLM workflows
the rest of the field is still Googling
21 modules. 110+ lessons. Full tidymodels + modern LLM stack. Six portfolio systems. 1 month of prebuilt VM access included.