🧭 AI & ML Vocabulary Hub
11 categories, 267 exercises. One map for every AI and machine learning English topic on Coders Lingo.
The AI/ML English landscape, in plain terms
Broadly, the ten categories below split into four groups. Foundational LLM vocabulary (AI & Prompt Engineering, Prompt Engineering Language) covers the terms you need the moment you start working with any large language model — tokens, context windows, hallucination, chain-of-thought. Building with AI (AI Agents Language, LLM Application Development, AI Code Generation Tools) covers the vocabulary of actually constructing agentic systems, LLM-powered features, and using AI coding assistants day to day. Operating and measuring AI (AI Model Evaluation, ML Model Serving, Machine Learning Language) covers how you know a model is good and how you run it reliably in production — spanning both classic ML and modern LLM systems. Finally, governance and specialised domains (AI Alignment & Safety, AI Ethics & Governance, Recommendation Systems) covers the policy, safety-research, and application-specific vocabulary that becomes relevant as AI systems scale and are deployed responsibly.
These categories are not duplicates of each other, even where the icons and topics look similar at a glance. Each card below states in one line exactly what makes that category distinct, so you can jump directly to the vocabulary you actually need rather than working through overlapping material.
The 10 AI & ML vocabulary categories
- Intermediate – Advanced
AI & Prompt Engineering
The entry point to AI vocabulary: writing precise prompts, evaluating LLM outputs, and the core RAG/embeddings/context-window terms every developer now needs.
Not a duplicate because: Broadest and most beginner-friendly set — start here if you are new to AI vocabulary.
- Intermediate – Advanced
Prompt Engineering Language
A deeper, more technical follow-on to AI & Prompt Engineering: system prompts, few-shot examples, chain-of-thought, temperature, and prompt-injection defence.
Not a duplicate because: Focuses on the engineering discipline of prompting — parameters, pipelines, and defensive techniques — rather than first-time vocabulary.
- Advanced
AI Agents Language
ReAct loops, tool use and function calling, agent memory, multi-agent orchestration, guardrails, and observability — the vocabulary of autonomous agents.
Not a duplicate because: Specific to agentic systems (agents that act, not just respond) — LangGraph, AutoGen, CrewAI, and the Claude Agents SDK vocabulary.
- Advanced
LLM Application Development Language
RAG pipeline vocabulary, function calling, LLM evaluation metrics, advanced prompt engineering, and LLMOps workflow — for engineers building LLM-powered products.
Not a duplicate because: Application-engineering vocabulary — how to build and operate LLM features in a product, rather than agent internals or prompt syntax alone.
- Intermediate – Advanced
AI Code Generation Tools Language
GitHub Copilot workflow, reviewing AI-generated code, Cursor IDE vocabulary, team adoption discussions, and writing prompts specifically for code generation.
Not a duplicate because: Narrowly about AI as a coding tool inside the IDE — a day-to-day developer topic distinct from building AI products.
- Advanced
AI Model Evaluation Language
Benchmarks, evaluation harnesses, faithfulness, recall, BLEU/ROUGE, and LLM-as-judge — the language for measuring and communicating model quality.
Not a duplicate because: Focused purely on measurement and evaluation vocabulary — the "how good is this model" conversation, distinct from serving or safety.
- Advanced
ML Model Serving
Serving architectures, model registries, inference optimisation, drift monitoring, shadow deployments, and explainability vocabulary.
Not a duplicate because: Production/MLOps vocabulary for running models in real systems — the operational side, not model quality or agent design.
- Advanced
Machine Learning Language
Explaining ML models to stakeholders, model evaluation vocabulary, model card writing, and experiment tracking language.
Not a duplicate because: Classic/traditional ML vocabulary (training, overfitting, feature engineering) rather than LLM-specific or generative AI terms.
- Advanced
AI Alignment & Safety Language
RLHF vocabulary (reward model, PPO), AI red-teaming language, alignment benchmarks, safety properties, and AI safety communication.
Not a duplicate because: Technical safety and alignment research vocabulary — how models are trained to behave safely, distinct from ethics/policy or evaluation.
- Advanced
AI Ethics & Governance Language
EU AI Act vocabulary, AI bias and fairness language, and AI governance communication exercises.
Not a duplicate because: Policy, law, and organisational governance vocabulary — the "should we" and "is this compliant" conversation, not technical training methods.
- Advanced
Recommendation Systems Language
Collaborative filtering, cold start, evaluation metrics, personalisation platforms, and fairness vocabulary — for ML and data engineers.
Not a duplicate because: A specific ML application domain (ranking and personalisation) with its own metrics and architecture vocabulary distinct from LLMs or generic ML.
Frequently asked questions
Why are there so many separate AI and ML vocabulary categories on Coders Lingo?
AI and machine learning English has grown into several genuinely distinct professional registers — the vocabulary a prompt engineer uses is not the vocabulary an MLOps engineer uses to discuss model serving, and neither overlaps much with the policy vocabulary of AI governance. Rather than force all of this into one oversized, unfocused category, Coders Lingo splits it into ten focused categories so each set of exercises stays specific and practical. This hub exists to help you find the right one quickly and see how they relate.
Which AI/ML category should I start with?
If you are new to AI vocabulary generally, start with AI & Prompt Engineering — it covers the foundational terms (LLM, RAG, embeddings, tokens, hallucination) that every other category assumes you already know. From there, branch out based on your role: developers using AI coding tools should go to AI Code Generation Tools Language; anyone building agentic systems should go to AI Agents Language; and ML/data engineers deploying models should go to ML Model Serving.
What is the difference between the "AI Agents" and "LLM Application Development" categories?
AI Agents Language focuses specifically on autonomous, multi-step systems — the ReAct loop, tool use, agent memory, and multi-agent orchestration. LLM Application Development Language is broader: it covers the vocabulary for building any LLM-powered feature, including RAG pipelines, function calling, and LLMOps, whether or not the system is agentic. Many engineers will use both categories together.
Is "Machine Learning Language" the same as the newer LLM-focused categories?
No. Machine Learning Language covers classic and traditional ML vocabulary — model training, overfitting, feature engineering, experiment tracking — the terminology used regardless of whether the model is a gradient-boosted tree or a neural network. The LLM-focused categories (AI & Prompt Engineering, Prompt Engineering Language, AI Agents, LLM Application Development) cover vocabulary specific to large language models and generative AI, which is a newer and more specialised vocabulary set.
Do AI Ethics & Governance and AI Alignment & Safety overlap?
They are related but distinct. AI Ethics & Governance Language covers policy, law, and organisational communication — EU AI Act vocabulary, bias and fairness discussions, and governance processes. AI Alignment & Safety Language covers the technical vocabulary used by researchers and engineers to train and evaluate models for safe behaviour — RLHF, red-teaming, and alignment benchmarks. A policy team member is more likely to need the former; a research or safety engineer is more likely to need the latter.
Should recommendation systems really be classified as an AI/ML vocabulary category?
Yes. Recommendation systems are one of the most widely deployed applications of machine learning, and the English used to discuss them — collaborative filtering, cold start, ranking versus retrieval, fairness in recommendations — is distinct enough from both classic ML and LLM vocabulary to warrant its own category. It is included in this hub because ML and data engineers working on personalisation need this vocabulary in addition to general ML terms.
How many total exercises are covered across the AI/ML vocabulary cluster?
The ten categories in this hub cover 267 exercises in total, ranging from beginner-friendly foundational vocabulary to advanced, research-level terminology. Each category is self-contained, so you do not need to complete them in order — pick the ones most relevant to your current role and work.
Is there a recommended order to work through the AI/ML vocabulary categories?
A practical order for most developers: (1) AI & Prompt Engineering for foundational vocabulary, (2) AI Code Generation Tools Language if you use Copilot, Cursor, or similar tools daily, (3) Prompt Engineering Language and AI Agents Language for deeper technical fluency, then (4) LLM Application Development, ML Model Serving, or Machine Learning Language depending on whether you build products, operate models in production, or work in classic ML. AI Model Evaluation, AI Alignment & Safety, AI Ethics & Governance, and Recommendation Systems are more specialised and can be studied whenever they become relevant to your work.
Explore more
Browse the full exercise library or the Data Scientist & ML Engineer guide for a structured reference.