85 saved links, 2017–2026 (peak: 2023). A tag with two lives. The first is classic NLP-as-a-field: chatbot primers in 2017, Stanford courses, the annotated Transformer. The second begins when LLMs swallowed the field — from 2023 on, nearly every save is natural language as the universal interface: to SQL, to Kubernetes, to the shell, to your cloud. NLP didn’t die; it became the front end of everything.
Related: LLMs · Machine Learning · Deep Learning · AI Agents
Learning NLP, the field
- the Hugging Face NLP course — “really a great resource to start with NLP and LLMs”; saved three times across three years.
- Stanford XCS224U: Natural Language Understanding and CMU Advanced NLP — attention and transformers (lectures).
- NLP for absolute beginners — Jeremy Howard’s high-signal Kaggle notebook.
- LLM University by Cohere — free, comprehensive, LLM-era NLP.
- the NLP-and-LLMs-from-scratch roadmap · 10 GitHub repos to master NLP · 30 implemented NLP projects.
- spaCy — the lean, fast alternative when an LLM is overkill.
- the NLTK book — the 2017 starting point, still on the shelf.
Transformers under the hood
- The Annotated Transformer — “Attention Is All You Need,” implemented line by line.
- tracing the transformer in diagrams — tokens to multi-head attention, visually.
- NLP illustrated: text encoding — how text becomes numbers.
- NLP Zero to Hero — basics to a transformer in 10 notebooks (Karpathy-inspired).
- the “granddaddy of LLMs” paper — the first attempt to do all of NLP with one neural network.
- Introduction to Information Retrieval — “never gets old,” and newly relevant in the RAG era.
Natural language as the interface
The tag’s second life — NL in, actions out.
- To SQL: NL2SQL with LangChain and LangSmith · Semantic Kernel natural-language-to-SQL with Azure OpenAI · querying SQL Server in natural language from PowerShell.
- To infrastructure: kubectl-ai — natural-language Kubernetes manifests · InfraCopilot, the NLP IaC editor · is natural-language IaC the future of cloud automation? · DevOpsGPT — requirements to working software.
- To the terminal: Shell AI — never write a shell command again · AI Shell Agent on LangChain ReACT.
- To your data: the Conversational Analytics API · the Google Analytics MCP server · Cosmos DB via natural language (MCP samples) · Copilot for Azure querying Cosmos DB (video).
- To the web itself: NLWeb — the Natural Language Web (video) · professional diagrams from natural-language descriptions · Playwright MCP — test generation from natural language.
- “natural language as a new, universal programming language” — Microsoft’s framing of the whole shift.
The newsletter thread
- Elvis Saravia’s NLP newsletter — the tag’s ongoing pulse; recent issues cover deep agents and the difference between “using a coding agent” and “engineering an autonomous coding system.”