This course focuses on fine-tuning and preference-tuning of LLMs for educational applications. Students will learn to build NLP/LLM pipelines that power modern educational technology, including content generation systems, automated assessment tools, conversational tutoring interfaces, and personalized learning platforms. They will gain hands-on experience with modern NLP architectures (transformers, LLMs) and training techniques (supervised fine-tuning, RLHF, DPO) for pedagogical alignment.
This course provides a comprehensive introduction to NLP through both foundational machine learning methods and modern neural architectures. Students begin by revisiting key ML concepts such as gradient descent, logistic regression, and neural network optimization before progressing to probabilistic models like Naive Bayes. The course then explores distributed word representations (Word2Vec), recurrent models (RNNs, LSTMs), and sequence-to-sequence attention architectures. Students also gain direct exposure to source code of transformer models. The course concludes with a deep dive into structured prediction methods using Hidden Markov Models, the Viterbi algorithm, and CKY parsing.