Glossary
The language of agentic AI, defined.
100 essential terms from foundational concepts to emerging technologies — your working reference for artificial intelligence terminology.
Generative AI and LLM Ecosystem
RAG (Retrieval-Augmented Generation)
Combines external data retrieval with generative models to improve accuracy.
Prompt Engineering
The art of crafting effective inputs to guide model outputs.
Token
Smallest unit of text processed by an LLM (roughly 4 characters or 0.75 words).
System Prompt
Hidden instruction guiding an AI model's overall behavior or persona.
LLM (Large Language Model)
AI trained on massive text datasets to generate human-like text.
Context Window
Maximum number of tokens a model can process in one prompt.
Hallucination
When a model generates false or fabricated information.
Vector Database
Specialized database for storing and searching embeddings.
Embeddings
Numeric vector representations of text, images, or audio used to measure similarity.
Prompt Injection
Malicious manipulation of model prompts to override its instructions.
Data Engineering and Pipelines
Feature Store
Centralized repository for ML features.
ELT (Extract, Load, Transform)
Variant optimized for modern data warehouses.
Schema
Structure defining how data is organized in a database.
Data Governance
Policies ensuring data accuracy, security, and compliance.
Data Warehouse
Structured repository optimized for analytics.
Data Lineage
Tracking the flow of data through systems and transformations.
ETL (Extract, Transform, Load)
Classic data pipeline pattern.
Data Quality Metrics
Measurements of completeness, accuracy, and consistency.
Data Lake
Raw data storage system for unstructured data.
Data Pipeline
Series of steps for ingesting, cleaning, transforming, and storing data.
Model Architectures and Math
Regularization
Prevents overfitting by adding constraints to model training.
Transformer
Neural architecture that underpins modern LLMs.
Encoder-Decoder Architecture
Used for translation and summarization tasks.
Diffusion Model
Generative model for images and video.
Gradient Descent
Optimization algorithm for training models.
GAN (Generative Adversarial Network)
Uses two neural nets competing to generate realistic outputs.
Loss Function
Quantifies how far predictions are from the target.
Attention Mechanism
Allows models to focus on relevant parts of input sequences.
Backpropagation
Algorithm for updating weights in neural networks.
Latent Space
Abstract vector space where model representations live.
Applied AI Techniques
Human-in-the-Loop (HITL)
Combining human oversight with automated AI systems.
Computer Vision (CV)
AI that processes and interprets visual data from images or video.
NLP (Natural Language Processing)
AI methods for understanding and generating human language.
Active Learning
Model queries humans for labeling the most informative samples.
Multimodal AI
Models that process text, images, and other modalities together.
Speech-to-Text (ASR)
Converting spoken language into written text.
Few-shot Learning
Learning from a very small amount of labeled examples.
Explainable AI (XAI)
Techniques to make model decisions interpretable.
Text-to-Speech (TTS)
Generating human-like voice from text.
Zero-shot Learning
Making predictions on unseen classes without direct training examples.
AI Infrastructure and Platforms
AutoML
Tools that automate model training and selection.
Model Registry
Central store for managing ML models and versions.
Edge AI
Running models directly on devices instead of the cloud.
Serving Layer
Infrastructure that delivers real-time predictions.
AIOps
Applying AI to IT operations and observability.
DataOps
Agile practices for data pipeline management.
CI/CD for ML
Continuous integration and delivery of ML code and models.
Inference
Running a trained model on new data to generate outputs.
MLOps
Operational framework for deploying and managing ML models.
Model Drift
When model performance degrades as data changes over time.
Security, Ethics and Compliance
Bias in AI
Systematic unfairness embedded in models.
AI Governance
Organizational oversight of AI model lifecycle and risks.
Fairness Metrics
Quantitative measures to detect and mitigate bias.
PII (Personally Identifiable Information)
Data that can identify an individual.
Model Watermarking
Techniques to verify model ownership or detect generated content.
Red Teaming
Stress-testing AI models for vulnerabilities or misuse.
Data Privacy
Protection of user information from unauthorized access.
GDPR / DPDP
Regulations governing personal data protection.
Responsible AI
Framework ensuring AI systems are ethical, safe, and transparent.
Adversarial Attack
Input designed to fool AI models.
AI Applications and Use Cases
Recommender System
Suggests items based on user preferences.
Agentic AI
Autonomous systems that can plan and act toward goals.
Chatbot
Conversational AI system for automated dialogue.
Predictive Analytics
Using historical data to forecast outcomes.
Anomaly Detection
Identifying unusual behavior or data points.
Knowledge Graph
Network of relationships among entities for reasoning.
Copilot
AI assistant that augments human work.
Document Intelligence
Extracting structured data from documents.
Synthetic Data
Artificially generated data used for model training.
Semantic Search
Search using meaning rather than exact keywords.
Foundational AI Concepts
Foundation Model
Large pre-trained model (e.g., GPT, Claude, Gemini) that can be adapted to many downstream tasks.
Fine-tuning
Adapting a pre-trained model for a specific use case.
Deep Learning (DL)
Subset of ML using neural networks with multiple layers to extract higher-level features.
Machine Learning (ML)
Algorithms that learn patterns from data without explicit programming.
Neural Network
Computational model inspired by the human brain, consisting of nodes (neurons) and layers.
Supervised Learning
ML approach using labeled data to train models.
Artificial Intelligence (AI)
Systems that simulate human intelligence processes such as learning, reasoning, and problem-solving.
Transfer Learning
Using a pre-trained model as a starting point for a new task.
Reinforcement Learning (RL)
Models learn through trial and error by receiving rewards or penalties.
Unsupervised Learning
ML approach where the system identifies patterns in unlabeled data.
Tools, APIs and Frameworks
LangChain
Framework for building LLM applications with memory, tools, and data retrieval.
Weights & Biases (W&B)
Tool for experiment tracking and model management.
Amazon Bedrock
AWS service offering access to multiple foundation models.
Google Vertex AI
Google Cloud's managed AI platform.
OpenAI API
Interface to access GPT models.
Anthropic Claude
LLM known for safety and long-context reasoning.
TensorFlow
Deep learning framework by Google.
PyTorch
Widely used ML framework developed by Meta.
Hugging Face
Platform for hosting and sharing open-source models.
LlamaIndex
Framework for indexing data sources for RAG systems.
Advanced and Emerging Topics
Tool Use (Function Calling)
Allowing models to interact with APIs and data sources.
Quantization
Compressing models by reducing precision for faster inference.
Chain of Thought (CoT)
Step-by-step reasoning method in LLMs.
AutoGen / CrewAI
Frameworks for orchestrating multiple AI agents collaboratively.
Distillation
Training a smaller model to replicate a larger one's performance.
Sovereign AI
Building AI models and infrastructure locally to retain data control and compliance.
Agent Frameworks
Toolkits for building multi-step AI agents.
LoRA (Low-Rank Adaptation)
Efficient fine-tuning technique for large models.
Tree of Thoughts (ToT)
Structured multi-path reasoning for decision-making.
Multimodal Fusion
Integrating multiple data types (text, image, audio) in one model.
Ready to put these concepts into practice?
Agents run the work. Your team stays in control where it matters.