AI Glossary
A one-stop guide to core concepts in enterprise AI. Whether you're a business decision-maker evaluating AI solutions or a developer seeking deeper technical understanding, you'll find the definitions you need here.
📊AI Data Analytics
NL2SQL (Natural Language to SQL)
Technology that converts natural language into SQL queries. Users describe their data analysis needs in plain language, and the system automatically generates SQL statements. This is the core technology of AI data analytics, dramatically lowering the technical barrier to data querying and enabling non-technical users to derive insights directly from databases.
Semantic Modeling Layer (MDL)
A business semantic abstraction layer built on top of physical databases, mapping table and field names to business-friendly terms (e.g., mapping 'customer_ltv' to 'Customer Lifetime Value'). The semantic modeling layer is critical infrastructure for NL2SQL accuracy.
Multi-Dimensional Analysis Cube
A pre-computed multi-dimensional data aggregation structure that enables rapid analysis of business metrics across multiple dimensions (time, region, product, etc.). AI analytics platforms use Cube technology for sub-second streaming responses, even with billions of data rows.
SSE Streaming (Server-Sent Events)
Technology for pushing real-time data streams from server to client. In AI data analytics, SSE enables progressive result delivery — users see analysis progress without waiting for complete results, similar to ChatGPT's word-by-word output experience.
Vector Database
A database system specifically designed for storing and retrieving high-dimensional vector data. In AI data analytics, vector databases power semantic memory systems — storing natural language queries and historical SQL mappings as vectors, enabling increasingly accurate results through similarity matching.
🤖AI Agent
AI Agent (Artificial Intelligence Agent)
An AI system capable of autonomously perceiving its environment, making decisions, and executing actions. Unlike passive chatbots, AI Agents are proactive — they can invoke tools, execute multi-step tasks, and adjust strategies based on results. Enterprise AI Agents can automatically handle tickets, generate reports, and monitor systems.
Docker Containerization
Using Docker container technology to package applications and their dependencies into standardized, portable units. In AI Agent management, containerization enables instant creation, environment isolation, and resource management of Agent instances, ensuring different tenants' Agents don't interfere with each other.
Agent Runtime
The underlying execution environment for AI Agents, providing model invocation, tool integration, and state management capabilities. Common Agent runtimes include OpenClaw, Claude Code, and Hermes, each with different strengths in performance, cost, and scalability.
Token-Based Billing
A billing model based on the actual number of tokens processed by a large language model. A token roughly equals one Chinese character or 0.75 English words. Input tokens (prompts) and output tokens (generated content) are typically billed separately, enabling enterprises to precisely control AI usage costs.
Sandbox Isolation
A security mechanism that restricts running programs to a controlled environment, preventing them from accessing or affecting other parts of the system. In AI Agent platforms, each Agent instance runs in an isolated Docker container, ensuring they don't interfere with each other and preventing security risk propagation.
🧠LLM & AI Fundamentals
LLM (Large Language Model)
Massive neural network models trained on vast amounts of text data, with powerful language understanding and generation capabilities. Notable LLMs include GPT, Claude, DeepSeek, Qwen, and others. Enterprises access LLMs via APIs for intelligent Q&A, text analysis, code generation, and more.
RAG (Retrieval-Augmented Generation)
A technical architecture combining information retrieval with LLM generation — first retrieve relevant information from a knowledge base, then provide the results as context for the LLM to generate answers. RAG effectively reduces hallucination and grounds AI responses in real enterprise data, making it a core pattern for enterprise AI applications.
Prompt Engineering
The practice of designing and optimizing input prompts to guide LLMs toward more accurate and useful outputs. Effective prompt engineering — including role setting, few-shot examples, and chain-of-thought guidance — significantly improves AI response quality. In enterprise scenarios, prompt engineering directly impacts AI application effectiveness.
Fine-Tuning
The process of further training a pre-trained LLM on domain-specific labeled data. Fine-tuning enables general-purpose LLMs to better understand industry terminology and follow specific format requirements. Compared to training from scratch, fine-tuning is low-cost and fast, making it an important technique for vertical AI applications.
Multimodal AI
AI systems capable of simultaneously understanding and processing multiple data types (text, images, audio, video, etc.). Multimodal AI can answer questions about images, recognize tabular data in pictures, understand voice commands, and more, providing richer interaction modes and application scenarios for enterprises.
🏢Enterprise AI
RBAC (Role-Based Access Control)
A mechanism that manages system access permissions based on a user's role within an organization. In enterprise AI platforms, RBAC ensures analysts can only view their department's data and Agent admins can only manage their authorized Agent instances — foundational to multi-tenant security.
On-Premises Deployment
Deploying software on an enterprise's own servers or private cloud rather than the SaaS provider's public cloud. On-premises deployment ensures data stays entirely within the corporate network, meeting data security and compliance requirements for heavily regulated industries like finance and healthcare.
GEO (Generative Engine Optimization)
Content optimization strategies for AI-powered generative search engines (ChatGPT Search, Perplexity, etc.). Unlike traditional SEO, GEO focuses on structured data, citation authority, and content citability to help brand content be cited and recommended in AI-generated answers.
API-First Architecture
A design philosophy that treats APIs as the core of a product — first design unified, comprehensive API interfaces, then build web interfaces and mobile apps on top. API-first architecture enables enterprise AI products to easily integrate with other systems (CRM, ERP, OMS, etc.) for automated data and workflow connectivity.
SLA (Service Level Agreement)
A formal commitment between a service provider and customer regarding service quality, typically including availability (e.g., 99.9%), response time, and recovery time. Enterprise AI platforms use SLA management to ensure critical tickets are handled within specified timeframes, with automatic escalation on timeout.
Ready to Apply AI in Your Enterprise?
From concept to implementation, YituTech helps you find the right AI solution