Tutorials

This tutorial series starts with executable Jupyter notebooks for one library at a time, then composes the libraries around reproducible design-research workflows. Every notebook renders its saved results directly after the cell that produced them. Only the Agents propose/critic notebook and final local-model walkthrough require a running model service.

Choose A Starting Point

Install design-research to follow the complete series. Install a component package directly when you only need its focused notebook. Each notebook page includes its exact environment command and a source download; see Run An Example In VS Code for the complete VS Code setup.

Tutorial paths

Tutorial	Primary skill	Libraries
Problems: Map the Word-Problem Catalog with t-SNE	Map 126 packaged word problems with TF-IDF and t-SNE.	Problems
Problems: Apply a Planar-Truss Grammar by Hand	Inspect and manually apply planar-truss grammar rules.	Problems
Agents: Refine a Design Rationale with Propose/Critic	Refine a design rationale with an existing LLM pattern.	Agents, Ollama
Agents: Build a Deterministic Workflow	Build a deterministic workflow from logic steps.	Agents
Experiments: Test the Monty Hall Strategy	Test stay versus switch as a seeded experiment.	Experiments
Analysis: Measure Inter-Rater Reliability	Quantify agreement among protocol coders.	Analysis
Compose The Libraries: Run A Benchmark Study	Run one benchmark through canonical artifacts.	All four libraries
Compose The Libraries: Compare Design Processes	Compare condition-specific design-process traces.	Problems, Experiments, Analysis
Compose The Libraries: Analyze A Partial Factorial Study	Fit a regression from a partial factorial study.	Problems, Experiments, Analysis

Recommended Order

1. Start with either Problems notebook, then compare the existing Agents pattern with the home-built workflow.

2. Complete the Monty Hall and reliability notebooks.

3. Run Compose The Libraries: Run A Benchmark Study to see the shared artifact contract.

4. Choose Compose The Libraries: Compare Design Processes for sequence analysis or Compose The Libraries: Analyze A Partial Factorial Study for design-of-experiments analysis.

5. Continue to Prompt Strategy Comparison Walkthrough when a local llama.cpp model should replace the deterministic agent.

All runnable files live under examples/. make examples-test executes the complete offline set. Set RUN_OLLAMA_EXAMPLES=1 to include the propose/critic notebook after starting Ollama. Set RUN_LLAMA_CPP_EXAMPLES=1 for the separate managed llama.cpp walkthrough; the two selections are independent. make notebooks-check verifies that every focused notebook page still shows results corresponding to its committed source.