AI & ML Glossary
Key terms and concepts in AI engineering, explained for practitioners. Each entry includes salary impact data and related skills.
AI moves fast and the terminology moves faster. This glossary covers the core concepts you'll encounter in AI engineering job postings, interviews, and technical discussions, from foundational architectures like transformers to applied techniques like RAG and fine-tuning. Each entry explains how the technology works, why it matters for your career, and what salary premium it commands in the current job market.
All Terms
AI systems that autonomously plan, execute, and iterate to accomplish multi-step goals using LLMs combined with tools, m...
AI Bias (AI Bias)Systematic errors or unfair patterns in AI model outputs that disadvantage certain groups, perspectives, or use cases. A...
AI Hallucination (AI Hallucination)When an AI model produces output that is plausible-sounding but factually incorrect or fabricated. Hallucination is one ...
AI Safety (AI Safety & Alignment)The field of research and engineering focused on ensuring AI systems behave as intended, avoid harmful outputs, and rema...
Agentic AI (AI Agents)AI systems that can autonomously plan, reason, and execute multi-step tasks. Agents use LLMs as reasoning engines combin...
Chain of Thought (Chain of Thought (CoT))A prompting technique that asks an LLM to produce intermediate reasoning steps before providing a final answer. Chain of...
Computer Vision (Computer Vision)A field of AI that enables machines to interpret and understand visual information from images and video. Computer visio...
Constitutional AI (Constitutional AI)An approach to training AI assistants developed by Anthropic that uses AI feedback guided by a set of principles (a "con...
Context Window (Context Window)The maximum number of tokens an LLM can process in a single request, including both the input prompt and the generated o...
CrewAI (CrewAI)An open-source framework for orchestrating multiple AI agents that collaborate to complete complex tasks. CrewAI enables...
DPO (Direct Preference Optimization)A technique for aligning LLMs with human preferences without the complexity of reinforcement learning. DPO has largely r...
Diffusion Models (Diffusion Models)A class of generative AI models that create data (images, audio, video) by learning to reverse a gradual noising process...
Embeddings (Vector Embeddings)Numerical representations of text, images, or other data in high-dimensional vector space. Embeddings capture semantic m...
Evals (AI Evaluations)Frameworks and methodologies for measuring AI model and application quality. Evals are the production-grade equivalent o...
Feature Store (Feature Store)A centralized platform for storing, managing, and serving machine learning features. Feature stores ensure consistent fe...
Few-Shot Learning (Few-Shot Learning)A prompting technique where the LLM is shown a small number of input-output examples before being asked to handle a new ...
Fine-tuning (Model Fine-tuning)The process of training a pre-trained LLM on a smaller, domain-specific dataset to adapt it for particular tasks or indu...
Flash Attention (Flash Attention)An optimized algorithm for computing transformer attention that reduces memory usage and increases throughput. Flash Att...
Foundation Model (Foundation Model)A large AI model trained on broad data that can be adapted to a wide range of downstream tasks. GPT-4, Claude, Gemini, a...
Function Calling (Function Calling (Tool Use))A capability of modern LLMs to call external functions or tools as part of generating a response. Function calling enabl...
Inference Optimization (LLM Inference Optimization)Techniques for making AI model inference faster and cheaper in production, including quantization, batching, caching, di...
Jailbreaking (AI Jailbreaking)Techniques to bypass an AI model's safety guardrails, causing it to produce content the model is trained to refuse. Jail...
KV Cache (KV Cache (Key-Value Cache))A memory optimization that caches the key and value tensors from previous tokens during LLM generation, avoiding redunda...
Knowledge Graph (Knowledge Graph)A structured representation of facts and relationships between entities, stored as a network of nodes and edges. Knowled...
LLM (Large Language Model)A neural network trained on massive text datasets that can generate, understand, and reason about text. LLMs like GPT-4,...
LangGraph (LangGraph)A framework built on top of LangChain for building stateful, multi-actor AI agent applications using graph-based workflo...
LangSmith (LangSmith)LangChain's observability and evaluation platform for LLM applications. LangSmith provides tracing, debugging, and eval ...
LoRA (Low-Rank Adaptation)A parameter-efficient fine-tuning technique that adds small trainable matrices to a frozen pre-trained model, enabling a...
MLOps (Machine Learning Operations)The set of practices for deploying, monitoring, and maintaining ML models in production. MLOps combines ML, DevOps, and ...
Mixture of Experts (Mixture of Experts (MoE))A neural network architecture that splits a model into specialized sub-networks (experts), with a gating mechanism routi...
Model Distillation (Model Distillation)A technique that trains a smaller "student" model to imitate the behavior of a larger "teacher" model. Distillation prod...
Model Registry (Model Registry)A centralized repository for storing, versioning, and managing machine learning models throughout their lifecycle. Model...
Multimodal AI (Multimodal Artificial Intelligence)AI systems that can process and generate multiple types of data simultaneously, such as text, images, audio, and video. ...
NLP (Natural Language Processing)A branch of AI focused on enabling computers to understand, interpret, and generate human language. NLP encompasses ever...
PEFT (Parameter-Efficient Fine-Tuning)A family of techniques that fine-tune large language models by updating only a small subset of parameters, dramatically ...
Positional Encoding (Positional Encoding)A method for injecting position information into transformer inputs, since self-attention is permutation-invariant. Mode...
Prompt Caching (Prompt Caching)A capability of modern LLM APIs to cache portions of prompts (typically system prompts and few-shot examples), reducing ...
Prompt Engineering (Prompt Engineering)The practice of designing and optimizing input prompts to get desired outputs from large language models. It involves un...
Prompt Injection (Prompt Injection)A class of attacks where malicious input manipulates an LLM into ignoring its system prompt or producing unintended beha...
Quantization (Model Quantization)A technique for reducing the precision of model weights from 32-bit or 16-bit floating point to smaller formats (8-bit, ...
RAG (Retrieval-Augmented Generation)A technique that enhances LLM responses by retrieving relevant documents from a knowledge base before generating an answ...
RLHF (Reinforcement Learning from Human Feedback)A training technique where human preferences are used to guide model behavior. Evaluators rank model outputs, and this f...
Reasoning Models (Reasoning Models)AI models specifically trained to perform extended reasoning through internal chain of thought. OpenAI's o1 series, Deep...
Reranker (Reranker)A model that takes a set of retrieved documents and reorders them by relevance to the query. Rerankers significantly imp...
Self-Attention (Self-Attention Mechanism)The core operation in transformer architectures where each position in a sequence attends to every other position, compu...
Semantic Search (Semantic Search)Search that understands the meaning of queries rather than just matching keywords. Semantic search uses embeddings and v...
Speculative Decoding (Speculative Decoding)An inference optimization that uses a smaller "draft" model to generate candidate tokens, which a larger model then veri...
Test-Time Compute (Test-Time Compute)The amount of computation an AI model uses during inference (response generation) rather than during training. Modern re...
Tokenization (Tokenization)The process of breaking text into smaller units (tokens) that LLMs can process. Tokens are typically subword pieces, not...
Tool Use (Tool Use)A capability of LLMs to invoke external functions, APIs, or tools as part of generating a response. Tool use is the foun...
Transformers (Transformer Architecture)The neural network architecture behind modern LLMs. Transformers use self-attention mechanisms to process sequences in p...
Vector Database (Vector Database)A database designed to store, index, and query high-dimensional vector embeddings efficiently. Vector databases enable s...
World Models (World Models)AI systems that learn an internal representation of an environment's dynamics, allowing them to plan, predict, and reaso...
Zero-Shot Learning (Zero-Shot Learning)Asking an LLM to perform a task without providing any examples, relying solely on the task description in the prompt. Ze...