Sign inStart free
AI Engineering/Phase 05
Core

NLP Foundations to Advanced

Tokenization to sequence models: how machines read language.

29 lessons ~30.5 hr
  1. Text Processing — Tokenization, Stemming, LemmatizationBuild · Free · 45 minLanguage is continuous. Models are discrete. Preprocessing is the bridge.Open
  2. Bag of Words, TF-IDF, and Text RepresentationBuild · 1 h 15 minCount first, think later. TF-IDF still beats embeddings on well-defined tasks in 2026.Needs Full Access
  3. Word Embeddings — Word2Vec from ScratchBuild · 1 h 15 minA word is the company it keeps. Train a shallow net on that idea and geometry falls out.Needs Full Access
  4. GloVe, FastText, and Subword EmbeddingsBuild · 45 minWord2Vec trained one embedding per word. GloVe factorized the co-occurrence matrix. FastText embedded the pieces. BPE bridged to transformers.Needs Full Access
  5. Sentiment AnalysisBuild · 1 h 15 minThe canonical NLP task. Most of what you need to know about classical text classification shows up here.Needs Full Access
  6. Named Entity RecognitionBuild · 1 h 15 minPull the names out. Sounds easy until you deal with ambiguous boundaries, nested entities, and domain jargon.Needs Full Access
  7. POS Tagging and Syntactic ParsingBuild · 45 minGrammar was unfashionable for a while. Then every LLM pipeline needed to validate structured extraction, and it came back.Needs Full Access
  8. CNNs and RNNs for TextBuild · 1 h 15 minConvolutions learn n-grams. Recurrences remember. Both are superseded by attention. Both still matter on constrained hardware.Needs Full Access
  9. Sequence-to-Sequence ModelsBuild · 1 h 15 minTwo RNNs pretending to be a translator. The bottleneck they hit is the reason attention exists.Needs Full Access
  10. Attention Mechanism — The BreakthroughBuild · 45 minThe decoder stops squinting at a compressed summary and starts looking at the whole source. Everything after this is attention plus engineering.Needs Full Access
  11. Machine TranslationBuild · 1 h 15 minTranslation is the task that paid for NLP research for thirty years and keeps paying now.Needs Full Access
  12. Text SummarizationBuild · 1 h 15 minExtractive systems tell you what the document said. Abstractive systems tell you what the author meant. Different tasks, different pitfalls.Needs Full Access
  13. Question Answering SystemsBuild · 1 h 15 minThree systems shaped modern QA. Extractive found spans. Retrieval-augmented grounded them in documents. Generative produced answers. Every modern AI assistant is a mix of the three.Needs Full Access
  14. Information Retrieval and SearchBuild · 1 h 15 minBM25 is precise but brittle. Dense casts a wide net but misses keywords. Hybrid is the 2026 default. Everything else is tuning.Needs Full Access
  15. Topic Modeling — LDA and BERTopicLearn · 45 minLDA: documents are mixtures of topics, topics are distributions over words. BERTopic: documents cluster in embedding space, clusters are topics. Same goal, different decompositions.Needs Full Access
  16. Text Generation Before Transformers — N-gram Language ModelsBuild · 45 minIf a word is surprising, the model is bad. Perplexity makes surprise a number. Smoothing keeps it finite.Needs Full Access
  17. Chatbots — Rule-Based to Neural to LLM AgentsLearn · 1 h 15 minELIZA replied with pattern matches. DialogFlow mapped intents. GPT answered from weights. Claude runs tools and verifies. Each era solved the previous one's worst failure.Needs Full Access
  18. Multilingual NLPLearn · 45 minOne model, 100+ languages, zero training data for most of them. Cross-lingual transfer is the practical miracle of the 2020s.Needs Full Access
  19. Subword Tokenization — BPE, WordPiece, Unigram, SentencePieceLearn · 1 hWord tokenizers choke on unseen words. Character tokenizers blow up sequence length. Subword tokenizers split the difference. Every modern LLM ships on one.Needs Full Access
  20. Structured Outputs & Constrained DecodingBuild · 1 hAsk an LLM for JSON. Get JSON most of the time. In production, "most" is the problem. Constrained decoding turns "most" into "always" by editing the logits before sampling.Needs Full Access
  21. Natural Language Inference — Textual EntailmentLearn · 1 h"t entails h" means a human reading t would conclude h is true. NLI is the task of predicting entailment / contradiction / neutral. Boring on the surface, load-bearing in production.Needs Full Access
  22. Embedding Models — The 2026 Deep DiveLearn · 1 hWord2Vec gave you a vector per word. Modern embedding models give you a vector per passage, cross-lingual, with sparse, dense, and multi-vector views, sized to fit your index. Pick wrong and your RAG retrieves the wrong thing.Needs Full Access
  23. Chunking Strategies for RAGBuild · 1 hChunking configuration influences retrieval quality as much as the choice of embedding model (Vectara NAACL 2025). Get chunking wrong and no amount of reranking saves you.Needs Full Access
  24. Coreference ResolutionLearn · 1 h"She called him. He did not answer. The doctor was at lunch." Three references to two people and nobody is named. Coreference resolution figures out who is who.Needs Full Access
  25. Entity Linking & DisambiguationBuild · 1 hNER found "Paris." Entity linking decides: Paris, France? Paris Hilton? Paris, Texas? Paris (the Trojan prince)? Without linking, your knowledge graph stays ambiguous.Needs Full Access
  26. Relation Extraction & Knowledge Graph ConstructionBuild · 1 hNER found the entities. Entity linking anchored them. Relation extraction finds the edges between them. A knowledge graph is the sum of nodes, edges, and their provenance.Needs Full Access
  27. LLM Evaluation — RAGAS, DeepEval, G-EvalBuild · 1 h 15 minExact-match and F1 miss semantic equivalence. Human review does not scale. LLM-as-judge is the production answer — with enough calibration to trust the number.Needs Full Access
  28. Long-Context Evaluation — NIAH, RULER, LongBench, MRCRLearn · 1 hGemini 3 Pro advertises 10M tokens of context. At 1M tokens, 8-needle MRCR drops to 26.3%. Advertised ≠ usable. Long-context evaluation tells you the actual capacity of the model you are shipping on.Needs Full Access
  29. Dialogue State TrackingBuild · 1 h 15 min"I want a cheap restaurant in the north... actually make it moderate... and add Italian." Three turns, three state updates. DST keeps the slot-value dict in sync so the booking works.Needs Full Access