← Back to Curriculum

Week 6: LLM Engineering

Days 36–42 · 17.5 hours

This week covers the engineering side of LLMs: evaluation, quantization, in-context learning, long context, RAG, tool use, and applying LLMs to robotics. Culminates in Phase III Capstone Day 1.

Daily Lessons

Day Topic Focus
36 LLM Evaluation Perplexity, MMLU, HumanEval, LLM-as-judge
37 Quantization & Inference INT4/INT8, GPTQ, AWQ, vLLM
38 In-Context Learning Zero/few-shot, mesa-optimization
39 Long Context & Reasoning RoPE scaling, ring attention, o1-style
40 RAG & Tool Use Retrieval-augmented generation, function calling
41 LLM for Robotics SayCan, Code as Policies, fleet planning
42 Phase III Capstone Day 1 Fine-tune robotics assistant + RAG

Key Concepts

  • Evaluation: How to measure if an LLM is actually good — benchmarks, contamination, Chatbot Arena
  • Quantization: Compress 7B models to run on consumer hardware with minimal quality loss
  • In-context learning: The most surprising emergent ability — learning from examples in the prompt
  • Long context & reasoning: Scaling context windows and chain-of-thought for complex tasks
  • RAG: Augment LLMs with external knowledge without fine-tuning
  • LLMs for robotics: From language understanding to physical world actions

Study Notes References