Curated notes, notebooks, and small demos covering LLM concepts, patterns, and agent architectures. Designed for quick learning and reproducible experiments.
- Requirements: Python 3.11 (see
.python-version).
- Setup (recommended):
python -m venv .venv && source .venv/bin/activate
pip install jupyter
jupyter lab # or: jupyter notebook
- Execute a notebook headlessly (smoke test):
jupyter nbconvert --to notebook --execute path/to/notebook.ipynb --inplace
| Topic |
Description |
Link |
| AI Engineering vs ML Engineering |
Comparison of AI and ML engineering approaches |
Link |
| Agents |
Implementation and concepts of LLM agents |
Link |
| More than just a next token predictor |
Deeper exploration of LLM capabilities |
Link |
| Productionizing an ML system |
Guide for deploying ML systems to production |
Link |
| Topic |
Description |
Link |
| Function Calling |
Implementing function calling with LLMs |
Link |
| Structured Output |
Emitting JSON and schemas from LLMs |
Link |
| Structured Extraction |
Extracting structured data with LLMs |
Link |
| Structured Extraction with Llama |
Using Llama for data extraction |
Link |
| Record Linkage |
Entity resolution and record matching with LLMs |
Link |
| Semantic joins and record linkage |
Complete report: methods, NSO experiments, adverse results, costs and next actions |
Notebook |
| Pydantic |
Using Pydantic for data validation with LLMs |
Link |
| Topic |
Description |
Link |
| MCP notes |
Servers, clients, reference servers, resources, and prompts |
Link |
| Direct tool calling |
Call arXiv tools in the notebook process |
Link |
| MCP server |
Expose those tools over stdio |
Link |
| MCP client |
Connect one chatbot to the research server |
Link |
| Reference servers |
Connect one chatbot to several MCP servers |
Link |
| Prompts and resources |
Read-only paper data and a search prompt |
Link |
| Topic |
Description |
Link |
| LangGraph |
Small graph and agent experiments |
Link |
| DSPy |
DSPy notes and drafts |
Link |
| Prediction-powered inference |
Estimator notebooks |
Link |
| Topic |
Description |
Link |
| Attention |
Deep dive into attention mechanisms |
Link |
| Reinventing Attention |
First principles approach to attention |
Link |
| Prompting |
Techniques for effective prompting |
Link |
| Prompting LLMs |
Advanced prompting strategies |
Link |
| Sampling |
Sampling methods for text generation |
Link |
| Evaluation |
Metrics and methods for LLM evaluation |
Link |
| LLM Classification |
Using LLMs for classification tasks |
Link |
| Topic |
Description |
Link |
| Introduction |
Basics of reinforcement learning |
Link |
| Fundamentals |
Core RL concepts and algorithms |
Link |
| Advanced Topics |
More complex RL techniques |
Link |
| Applications |
Applying RL to practical problems |
Link |
| Policy Gradients |
Implementing and analyzing policy gradients |
Link |
| Additional Topics |
Extensions and case studies |
Link |
- Read the contributor guide:
agents.md.
- New docs, notebooks, and images should use kebab-case (e.g.,
structured-extraction.ipynb).
- Python modules should use snake_case to remain importable (e.g.,
research_server.py).
- Some notebooks are work-in-progress; run cells top-to-bottom and install missing libraries as prompted.
- Keep outputs lean to improve diffs and review.
