Natural Language Processing
Text preprocessing, word embeddings, and the path from RNNs to BERT — how machines learned to read.
8 lessons
01Text Preprocessing
Tokenization, BPE, stemming, lemmatization, TF-IDF — turning raw text into model-ready representations.
Word Embeddings
Word2Vec, GloVe, FastText — how dense vectors capture meaning, and why contextual embeddings supersede them.
Sequence Models to BERT
From seq2seq with attention to Transformers to BERT — how NLP went from task-specific to universal pre-training.
LLM Model Taxonomy
Encoder-only (BERT), decoder-only (GPT), and encoder-decoder (T5) families — what each was built for, and what 'large' actually means.
Decoding & Sampling Strategies
Turning a next-token distribution into actual text — greedy, beam search, temperature, top-k, top-p, repetition penalties, and the limits of determinism.
Topic Modeling with BERTopic
How modern topic modeling actually works — embed each document, reduce to a few dimensions, cluster, then label clusters with class-based TF-IDF. The four-stage pipeline that replaced LDA for most production text-mining workloads.
Training Embedding Models
How a generic transformer encoder gets fine-tuned into a sentence embedder — Sentence-BERT, contrastive (InfoNCE) and triplet losses, hard-negative mining, TSDAE for unsupervised pretraining, and Matryoshka embeddings for variable-size vectors.
NLP Quiz
Test your understanding of tokenization, embeddings, BERT, model families, decoding strategies, topic modeling, and embedding training.