Glossary

A

Agent - An autonomous system that perceives its environment and takes actions to achieve goals.

Agentic Framework - A software framework designed for building AI agents (e.g., LangChain, AutoGPT).

API (Application Programming Interface) - Interface for software components to communicate.

AST (Abstract Syntax Tree) - Tree representation of code structure.

B

Backoff - Strategy for retrying failed operations with increasing delays.

Benchmark - Standardized test for measuring performance.

Beam Search - Search algorithm that explores multiple paths simultaneously.

C

Chain-of-Thought (CoT) - Prompting technique that encourages step-by-step reasoning.

Checkpoint - Saved state of a model or agent for recovery.

Context Window - Maximum amount of text an LLM can process at once.

Constitutional AI - Approach to align AI behavior with principles.

D

Deterministic - Producing the same output given the same input.

Distributed Tracing - Tracking requests across multiple services.

Docker - Platform for containerizing applications.

E

Embedding - Vector representation of text or data.

Episodic Memory - Memory of specific past events or experiences.

Evaluation Metric - Quantitative measure of performance.

F

Few-Shot Learning - Learning from a small number of examples.

Fine-Tuning - Training a pre-trained model on specific data.

Function Calling - LLM capability to invoke external functions.

G

Generalization - Ability to perform well on unseen data.

Guardrails - Safety mechanisms to prevent harmful behavior.

GPU (Graphics Processing Unit) - Hardware for parallel computation.

H

Hallucination - When LLMs generate false or nonsensical information.

Human-in-the-Loop (HITL) - System requiring human approval for decisions.

Hyperparameter - Configuration parameter for model training.

I

Inference - Using a trained model to make predictions.

Interpretability - Ability to understand model decisions.

K

Kubernetes (K8s) - Container orchestration platform.

L

Latency - Time delay between request and response.

LLM (Large Language Model) - Neural network trained on vast text data.

Long-Horizon Planning - Planning over extended time periods.

M

Memory System - Component for storing and retrieving information.

Meta-Learning - Learning how to learn.

Microservices - Architecture pattern with independent services.

Multimodal - Processing multiple types of data (text, images, audio).

N

Neural Network - Computing system inspired by biological brains.

NLP (Natural Language Processing) - Processing and understanding human language.

O

Observability - Ability to understand system internal state from outputs.

Orchestration - Coordinating multiple components or agents.

P

Perception-Reasoning-Action Loop - Core agent cycle: observe, think, act.

Prompt Engineering - Crafting effective prompts for LLMs.

Production - Live environment serving real users.

R

RAG (Retrieval-Augmented Generation) - Combining retrieval with generation.

ReAct - Pattern combining reasoning and acting.

Reinforcement Learning (RL) - Learning through rewards and penalties.

RLHF (Reinforcement Learning from Human Feedback) - Training with human preferences.

S

Sandbox - Isolated environment for safe code execution.

Semantic Memory - Memory of facts and knowledge.

Semantic Search - Search based on meaning, not keywords.

Self-Improvement - Agent’s ability to improve its own capabilities.

Streaming - Sending responses incrementally as generated.

T

Temperature - Parameter controlling randomness in LLM outputs (0=deterministic, 1=creative).

Token - Unit of text processed by LLMs (roughly 0.75 words).

Tool - External function or API an agent can use.

Tree of Thoughts - Exploring multiple reasoning paths.

V

Vector Database - Database optimized for similarity search on embeddings.

Validation - Checking if outputs meet requirements.

W

Working Memory - Short-term memory for current task.

Z

Zero-Shot - Performing tasks without specific training examples.

Common Acronyms

• AI - Artificial Intelligence
• API - Application Programming Interface
• AST - Abstract Syntax Tree
• CI/CD - Continuous Integration/Continuous Deployment
• CoT - Chain-of-Thought
• GPU - Graphics Processing Unit
• HITL - Human-in-the-Loop
• LLM - Large Language Model
• ML - Machine Learning
• NLP - Natural Language Processing
• RAG - Retrieval-Augmented Generation
• RL - Reinforcement Learning
• RLHF - Reinforcement Learning from Human Feedback
• SLA - Service Level Agreement
• ToT - Tree of Thoughts
• UI/UX - User Interface/User Experience

Model Parameters

Temperature - Controls randomness (0.0-2.0)

• 0.0-0.3: Focused, deterministic
• 0.4-0.7: Balanced
• 0.8-1.0: Creative
• 1.0+: Very random

Top-p (Nucleus Sampling) - Alternative to temperature (0.0-1.0)

• 0.1: Very focused
• 0.5: Balanced
• 0.9: Diverse

Max Tokens - Maximum length of response

Frequency Penalty - Reduces repetition (-2.0 to 2.0)

Presence Penalty - Encourages new topics (-2.0 to 2.0)

HTTP Status Codes

• 200 - Success
• 400 - Bad Request
• 401 - Unauthorized
• 429 - Rate Limited
• 500 - Server Error
• 503 - Service Unavailable