A growing collection of practical, beginner-friendly projects using Python, LangChain, and LangSmith to explore modern LLM patterns.
Inspired by the LLM Engineering Cheatsheet
git clone https://github.com/mlane/llm-getting-started.git
cd llm-getting-started
python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.sample .env
# Fill in your OpenAI / LangSmith API keys
# Run an example project
python3.11 projects/debate_agent.py# Format code
black .
# Lint code
ruff check .This repo will grow over time. Projects are grouped by complexity to help you build intuition as LLM concepts evolve from simple to advanced.
✅ = Implemented & ready to run 💡 = Planned or conceptual for now
| Level | Project | Concepts Practiced | Status |
|---|---|---|---|
| 🟢 Beginner | Simple ChatBot with Memory | Interactive session, short-term memory | ✅ |
| 🟢 Beginner | Zero/Few-Shot Prompt Playground | Prompt patterns, zero-shot thinking | 💡 |
| 🟡 Intermediate | LLM Agent Debate | System prompts, disagreement modeling | ✅ |
| 🟡 Intermediate | Role-Based Support Assistant | Formatting, role control | 💡 |
| 🔴 Advanced | Retrieval QA from Local Docs | Vectorstores, retrieval chain | 💡 |
| 🔴 Advanced | Multi-Agent Task Planner | LangGraph, agent chaining | 💡 |
Each script lists one or more of the following concepts it demonstrates:
| Behavior | Concept |
|---|---|
| No examples given | zero-shot reasoning |
| Examples in prompt | few-shot prompting |
| Explicit persona or tone | persona control, system prompts |
| Responds to previous turns | conversation history, short-term memory |
| Simulates disagreement or debate | agent disagreement |
| Builds on previous answers | turn-based dialogue |
| Explains steps | chain of thought |
| Uses external data for context | RAG (Retrieval Augmented Generation) |
| Uses retrieved data to augment model | document retrieval, model augmentation |
| Structure/Flow | Concept |
|---|---|
| Live user input loop | interactive session |
| Two or more agents taking turns | multi-agent interaction |
| Message-based prompt passing | conversational flow |
| Uses tools or actions | tool execution |
| Semantic search for relevant context | document retrieval, semantic search, vectorization |
Use these as a reference when reading or extending scripts.
We believe the best way to learn LLMs is by doing — each script is small, focused, and teaches a core idea.
This repo is built to be:
- Modular
- Beginner-friendly
- Focused on thinking, not just syntax
- Updated as the LLM ecosystem evolves
PRs welcome. Please keep things clean, consistent, and low-dependency.