NLP, LLMs, and Agentic Systems: Transformers, RAG, Prompting

About the course

This course traces the evolution of NLP from classical statistical models to modern LLMs and the systems built on top of them. Students study transformer architecture in full detail, fine-tuning methods (fine-tuning, RLHF, PEFT), and how to build RAG systems and tool-using agents. It is a central course in the applied AI track and in NLP research.

What you'll learn

✓Implement a transformer from scratch, including scaled dot-product attention and positional encoding

✓Fine-tune pretrained LLMs for a specific task using PEFT methods (LoRA)

✓Build a RAG system with vector search, a reranker, and answer-quality evaluation

✓Design an agentic pipeline with tool use and context management

✓Evaluate LLM systems using a combination of automatic metrics and LLM-as-judge

Key topics

•Tokenization, word embeddings (Word2Vec, GloVe, BPE)

•Transformer architecture: self-attention, multi-head attention, positional encoding

•LLM pretraining: GPT, BERT, T5, LLaMA; scaling laws

•Fine-tuning: full fine-tuning, LoRA, QLoRA, prompt tuning (PEFT)

•RLHF, DPO, and alignment methods

•Retrieval-Augmented Generation (RAG): chunking, vector indexes, re-ranking

•Prompt engineering: chain-of-thought, few-shot, structured output

•Agentic systems: tool use, ReAct, multi-agent orchestration

•Evaluating LLMs: BLEU, ROUGE, BERTScore, LLM-as-judge

This description was generated automatically and has not yet been reviewed by an instructor — it's a draft for discussion.

About the course

What you'll learn

Key topics

Feedback