10. References

Glossary

Term	Definition
LLM	Large Language Model — neural network trained on vast text data
Transformer	Neural architecture using self-attention
Tokenization	Converting text to subword token IDs
Embedding	Dense vector representation of text
RAG	Retrieval-Augmented Generation — ground output in retrieved data
SFT	Supervised Fine-Tuning — training on labeled examples
QLoRA	Quantized Low-Rank Adaptation — efficient fine-tuning
Quantization	Reducing model precision (32-bit to 4-bit)
Agent	AI system that uses tools, memory, and planning autonomously

Cheatsheet

# Installation Cheatsheet
pip install ollama                    # Local LLMs
pip install langchain chromadb        # RAG
pip install transformers torch        # HuggingFace
pip install bitsandbytes peft trl     # Fine-tuning
pip install gradio                    # UI
pip install modal                     # Serverless

# Quick inference (HuggingFace)
from transformers import pipeline
pipe = pipeline("text-generation", model="gpt2")
pipe("Once upon a time")

# Quick RAG (LangChain + Chroma)
from langchain.chains import RetrievalQA
qa = RetrievalQA.from_chain_type(llm=llm, retriever=vectorstore.as_retriever())

# Quick fine-tuning (TRL)
from trl import SFTTrainer
trainer = SFTTrainer(model=model, train_dataset=dataset)
trainer.train()

Model Comparison

Model	Size	Best For	Access
GPT-4o	Frontier	General, multimodal, coding	API
Claude 3.5	Frontier	Analysis, writing, safety	API
Llama 3.2	1B-90B	Open-source, local deploy	Ollama / HF
Phi-3 / Phi-4	3.8B-14B	Efficient, edge devices	HF / Azure
DeepSeek V3 / R1	67B-671B	Reasoning, code, math	API / Ollama
Qwen 2.5	0.5B-72B	Multilingual, code	Ollama / HF

📘 Congratulations! You've completed the AI Engineer Core curriculum. You can now build LLM products, multi-modal chatbots, RAG pipelines, fine-tune models, and deploy multi-agent systems to production.